I love your sense of humor. You have a very clear understanding of the tech & you methodically dismantle the BS that so many pseudo-audio-experts spew. I record in 48k/24bits/chan, operate filters in 8x over-sample mode to minimize aliasing induced distortion. Final mix down to shaped-dither 48 or 44.1 in 16bits.
Finally another fantastic video about this topic besides Dan Warroll’s. We don’t think about needing to hear 96,000Hz frequencies when recording at 192k. What we are thinking about (in terms of auditory effect) is the blending of digital audio samples as they combine together in a virtual space. This “smoothing” effect becomes much more apparent on things like cymbals in drum overheads & instruments with longer envelopes that ring out. When combined with digital (or virtual) signal processing like high shelves, you are able to use more EQ with less apparent artifacting and aliasing when compared to lower resolutions. Some plugins compensate for this by upsampling inside the plugin for processing, then downsampling on the way out, but not all of them do, and I haven’t heard of a DAW doing this with their console strip (please correct if I’m mistaken). The argument listed here about CPU power becomes a bit more null & void as time goes on. 192kHz is also much better for any kind of time or pitch correction, since higher resolution gives more samples to stretch & blend. The last benefit I’d mention for 192kHz is the lower latency times on system buffers. On certain systems, this is extremely desirable and beneficial since desktop computer systems these days can handle the load with much greater ease than ever before.
This seems to come from the "stairstep" myth of representing samples. Sampling theorem states that everything within the bandlimited signal is captured . Thus, 192k does not capture anything within the audible range any better than 48k. This is easily provable with analog equipment as was demonstrated in this video ua-cam.com/video/cIQ9IXSUzuM/v-deo.htmlsi=KF5VEVHFq9ZM8885
"This “smoothing” effect becomes much more apparent on things like cymbals in drum overheads & instruments with longer envelopes that ring out." Try it - but before you get disappointed: The most well payed audio-engineers whose sole job it is to validate audio quality can not head a difference where there is none.
@@ABaumstumpf I’ve gone back & forth a million times on it for the past 20 years and found slight, but very noticeable differences in such cases. Not everything requires such high sample rates, but it can transform how all of the tracks work together on a mix, and ESPECIALLY when hard quantizing drums. Sounds more like you should try it, many, many times before relying on what someone else says about it.
@@JesterMasque Higher sample rates are massively important when doing any kind of intence time stretching in the digital domain. They can also help stop or reduce aliasing when you don't have oversampling avalible natively in a plug in that would benafit from it or in your DAW. If a top engineer can't hear aliasing in the audiable range, I'd question their position, and by using souly lower sample rates you will surely run into it if using diferant plug ins all the time.
In terms mastering from tape to CD, what makes most difference is: (1) the quality of the source tape (2) subtle changes to eq etc during transfer to bring "the breath of life" to the presentation (3) the quality of analogue to digital conversion (ADC). The limitations of early ADCs was recognized by Tony Faulkner, who modified Sonys for better performance & the team at Pacific Microsonics who were developing HDCD.
Was entertaining to listen to and glad that it’s pretty much factually accurate up till the SACD as really you missed how the supper high frequency low but rate encoding actually works and why the recording industry would want to use this for archival and data storage. Well worth listening and even subscribing to. 👍🏻
As an electronics engineer, I really enjoy our down to earth technical videos, keep them coming! Regarding DSD & SACD, it's no coincidence that Mobile Fidelity Sound Lab use it for their digital releases... From a hardware engineering perspective, a bit stream D to A is much simpler than the alternative so it's much easier to do it right.
I just love how you cleverly are able to push all of the right buttons, without actually having to push them. LOL Another very well-done and informative discussion sir!
Thank you for the video. Have you done a video on how the streaming services manipulate your music as far as loundness and compression when uploading your tracks to their platforms.
I have in mind at some point to compare one of my tracks on Spotify with the original master. Whether any difference will be heard through UA-cam's audio mangler will be an interesting question.
Thanks for this piece, I really enjoyed it. Having been trained as studio engineer back in the 80's, back when we were getting our heads round digital and just learning to edit without putting some tape on the cutting block, I feel I know many technical matters in regards to sound reproduction and recording techniques, but I learnt several things from watching this. In addition to my more formal training, I've also been an 'audiophile' since I bought my Roksan Xerxes turntable as a young bushy tailed 20 year old, and have spent the last 40 years enjoying fine home reproduction and keeping abreast of the developments in the industry. 40 years of this has clearly shown me there is still so much we dont know and dont know how to measure. Whilst I'm a scientist by nature and training, Science has always been about explaing the observable to me, not dismissing observations and so I'm happy to acknowledge that amplifiers sound different and that cables sound different. When its as clearly observable as I hear it I accept my experiences, guided by many years of both home and Studio Audio. I'm almost 60 now and I'd be lucky to hear anything much over 14Khz. But given a known decent digital mastering, I can wholeheartedly tell you I can hear a difference between a 24/96 and 24/192 version of the same master, and its not a small difference. This is not expectation bias as I can get my wife to swap between the recordings (easy with Roon and access to all the files via Qobuz or similar) and identify the recording sampling rate with very high accuracy. My wife just like a good sound, but she has regularly commented on some studio masters sounding astonishing too even though she does not undertsand why they are different. I cant explain why I can hear it so clearly, but I accept it. It is of course the mastering that makes the most difference to the quality of a recording and I can also point to many remastered Albums in 24/192 or 24/96 which sound much worse than an earlier master on a Japaneses CD at 16/44.1. With so much confusion and so much we still dont know, its no wander the industry is plauged by those peddling 'snake oil' and their very exsistance gives those of us happy to work with the observable, some resistance from those that confidently state 'if you cant measure it it doesnt exsist'. Its always been a conundrum for those of us who indulge in some audophilia whilst still seeking the scientific answers of how to measure what we can observe.
One subject that some people ask me at times but unfortunately i lack the skill to explain as comprehensively as you is what happens when you work in one bit/resolution rate and then export at another. Like what happens if you limit a project at 96khz/24bit and export at 44.1/16 and the other way around, what happens with the ISP etc etc. I think a video on the subject would be worth a lot
An correct optimal solution derives the sound wave using interpolation, e.g. cubic spline. Then it resamples the waveform to the new output format. In a lot of DAWs, at least Adobe Auditoon, if you zoom in to the max you’ll be shown both the samples and the interpolated waveform. There’s a lot of old misinformation about how very old stuff maybe worked with digital in the 80ies. According to that misinformation there’s a lot loss and strange dithering noise applied to counter act it. My friend tried to explain that voodoo stuff to me and I stared confused at him… Pretty sure no DAW in modern times uses strange voodoo bullshit. Calculating the waveform and resample the to the output format is the same approach with barely any up/down sampling issues. Why you’d ever go away from the waveform in any application that prioritizes accuracy would be beyond my comprehension. Maybe some hardware that requires absolute zero latency do weird voodoo still, but DAWs surely must just resample waveform at best accuracy possible.
@@randomgeocacher thank you for the elaborate answer. I’m asking because in the past somebody once told me that for one example, if you use a limiter work in one Bitrate , and then export to another, you might be ending up exporting the file without any limiting applied.I cannot, however remember the exact terminology what betray toward betrayed this applies
@@randomgeocacher A complication with doing things this way is that applying a brick-wall filter to a sampled signal may result in peaks which exceed the peak level of the original signal. As a simple example, if one has a signal which, after sampling at 1 million samples per second, looks like a 1000Hz square wave whose peak values are 0.01dB below saturation, and downsamples to 44.1kHz, each edge will have a certain amount of overshoot before the signal settles toward an equilibrium until the next edge approaches.
Audiophile William here, I've tried and tested many highend DAC's over the years, i never paid any attention to all the numbers just the sound. I borrowed a MSB Premier DAC that does 44.1kHz to 3,072kHz PCM up to 32 bits and from 1 to 8xDSD, in the end i decided to keep my old modified Audio Note DAC from 1998 which sounded just as good if not better than the MSB Premier. My old Audio note dac manual says it has no over sampling, no jitter reduction, no noise shaping and no re-clocking and uses the highest grade AD1865, 18bit stereo converter chip what ever that all means?? Audio Phil tickles me every time 🙂
Thanks for that , Audiophile William !! I don't feel bad now, about really liking my 2004 Apogee Mini DAC ! 😂. I think the clocks and analog sections of DACS have much more to do with sound quality rather than the DACS used .. Almost ALL DACS today are perfectly capable on specs ...
All modern "brand name" DACS like TI (formerly Burr Brown) are absolute overkill for "listening" to CD's. I recently purchased a Rotel RCD1572MKII that uses the TI PCM5252. However it's more than just a $10.00 chip to do a wonderful conversion. The OP amps and output side design and components are also extremely important. In my opinion most US$1,000+ CD players are more than satisfactory to most ears without needing an external DAC. @@LukeSchneiderEWI
I just listen to music and my latest CD player is even more "sensitive" to crappy recordings/mastering's than anything I've owned before. It is utterly merciless and as analytical as my B&W speakers are in a similar manner. @@stefanweilhartner4415
Even CD quality is "only" 44.1 kHz (16 bit), enabling almost 22kHz bandwidth. That is what 10 year olds can rarely hear, even at higher volumes, and only when not masked by lower frequencies. If you are above 25 years of age - forget everything above 18 kHz.
i am a hearing aid acoustic technician, 27 years old. in blind tests i can just about differentiate silence from 24 khz at 65 db. its more a feeling than a sound, like how a tense jaw has a "sound", but between your ears and higher. but at that point its not about any soecific data, but about if this noise has a natural harmonic with the music (not disonant, nit misding, not overbearing). that does not require actual info on the medium, but simply a harmonic overtone extrapolation with a nice roll off. so 48 khz recordings, which hits my differentiable 24 khz ears, and 2x supersampling + harmonic distortion ultrasound dsp work that rolls off up there, is undifferentiable to me from 96khz or above recordings. my ears have a 120db dynamic range between lowest detectable sound and uncomfortable volume, so 24 bit exceeds that range well at 144db dynamic range.
As a kid, I could hear the flyback transformer in old tube TV's. They operate at 30khz. I can still hear above 20khz in my 50's. Not all ears are created equally.
@@Chrisspru you state your ears have 120db dynamic range, as someone who works with ears, you will be well aware of the damage listening to 0db then a 120db peak will do to your ears, you will also be aware of the average DB in a room and how noise floor above this equates to perceivable dynamic range. The reality is, with a recording, all else equal, no human needs or can benefit from above 16.44.1 no human can perceive the differences, and even funnier is the lack of actual mics recording above 20Khz and the lack of audio equipment to actually have the capability to play a 120db dynamic range, never mind actually excerise the 96db range of 16/44.1 In professional studio work and mastering there is a point to 24 bit. But in HiFi its pretty much foo like cables to make more money.
@SastusBulbas1 my hearing goes from -10 db to 110 db. and people can tell the difference between the 16 bit and 24 bit in reproduction just barely with very good ears. a tiny noise difference, an odd feeling that hits 80% on differentiation in blind tests. its pretty weird. with the hz the difference is in high note accuracy in a similar odd feeling way
@@Chrisspru your hearing is the same standard of hearing every human has. People cannot tell the difference between 16 bit and 24 bit audio files, this has been done to death by studio engineers. What people do hear a difference in, is DAC's due to topology, and even differences in mastering. So far blind tests have shown that audiophiles cant even tell the difference between a cable and wire coat hanger, a Bryston or Alexis amp, and websites these days are full of information as to why this is, Audioholics and Audio Science Review can be quite blunt with measurements and reality vs feelings and imagination.
As audiophiles can hear what no mortal who have participated in scientific experiments can hear, perhaps they could donate their ears and brains to scientific research; preferably _ante mortem_ .
Brilliant! Ya old bloody bastard! Well done... Ya know it all comes down to a human ear. Just like modern TV's that claim a "billion" colors, when the human eye on average can only see about one million. The same applies here (or should I say "hear"). Keep doing your stuff. Love it...
I think the low noisefloor of 24 bit is convenient for recording but at the same time I think my music does require high sampling rates as well so 24/44.1 serves me well
What may seem crazy today might just be every day in the future. I’m not sure if this is a version of Moores law but I know my ears are not going to double in frequency response any time soon. Great humour with Audio Phil a supporting cast who knows his thang.
Thank you for the ever exilarating but serious presentation. I use, if I can choose, 48KHz sampling rate and 32bits floating point resolution (mostly to avoid internal clipping). And I export my masters in 24bit 48KHz AND 16bit 44.1 properly dithered. I have tried 96 KHz and I DON'T like it unless it is for Jazz or Classical recordings. For Pop and Rock I feel is losing some grit (It might be some sort of psich-acoustic impression on me knowing the sampling rate. In general I am fully satisfied with 48KHz and if I can use oversampling in som plugins for extra resolution in some cases I am fine. The most important fact is that if it sounds good (to me) then IT IS good and that's it.. And if in 30 years from now they are still reprinting my recordings, then, good luck. I don't think any sample rate difference will make a difference. But who knows... When I built my first home studio around an 3340S Teac in 1975 I was dreaming to have a way to have a full 24 track studio in my bedroom. And in 1992 I was dreaming about a software I named "de-blender" that was capable of separate the individual instruments from a store recording... And here we are... both are a reality now. How impressive. The next step is a software that mix the tracks with A.I. (hahah... ) But I hope not as good and as artful as myself. Thank you for your contribution to the demistification of sound engineering.
I had an SACD player and the sound stage it could throw out was stunning. No CD I had could do the same. Many years ago when single bit DACs started to hit the market I was on the hunt for a new CD player and it was easy to identify the single bit machines by the scale of the presentation. Don’t really know why but the high bit rate machines just seemed better.
The reason why you never see bitstream used in DAWs is they can't handle the processing that way. They would effectively have to convert it to PCM on the fly, process and then convert back to bitstream.
What if we took the single-bit sample to an effective infinite sample rate. We could use amplitude to modulate a single stylus moving across a spinning cylinder coated with wax so that the number of samples is limited only by the number of molecules in the wax.
High-frequency response would be limited by the inertia of the stylus. 1 bit sampling at a high frequency is essentially what a class D amplifier does.
The other point about filters that almost got talked about is that steeper (higher order) filters generally introduce unwanted phase shifts. I think a lot of people can notice this.
Broadcast audio is moving toward 48khz as the recording standard. Used to be 44.1 for a very long time. For playback, it's 48khz, but that is changing. The final link from the digital audio processing to the transmitter exciter is becoming 192khz, so that stereo multiplex and ancillary data can ge generated by the computer doing the processing.
With regards YT's algorithms, I'm not so sure they're clever at all. They show videos on similar topics to the ones that I've watched recently but... they also persist in recommending items about Class 55 Deltic locomotives that my son used to love. Love that is... 10 sodding years ago! BTW... loving your channel :)
That makes sense to me :) also if you pair with a specialty small condenser mic that can capture beyond human hearing range (there’s a few that aren’t crazy expensive) you could even recover human inaudible higher frequencies. In theory it could be cool to hear animals like bats well, or recover unheard ranges of chaotic sounds like explosions or crashing glassware..,
My sampled instruments are 24/48 so I don't bother using anything other. Equipment is important and ensuring you have good AD-DA conversion and monitoring equipment. As for venue recording - keep it simple. Good tear down of the matter.
96kHz is very popular in live sound, as it halves signal propagation (processing) delay, which improves fold back to performers. Regarding DSD & SACD, to my aged, & admittedly rather abused, hearing; I find it sounds far more natural, open & clean (best words I can think of) but that might just be down to how they were recorded & mixed and that they have not suffered from being limited to death as is current CD mixing & mastering practice.
Another DSD/SACD enthusiast like me. I agree with you wholeheartedly regarding the cleanness and openness of the sound. I also find it true with regard to transients such as cymbals and drum thwacks, Keyboard fingering on the piano is more evident too. I have been downloading 5.6448MHz DSF (DSD128) albums from NativeDSD Music for some time now and most of them sound wonderful.
Native DSD has much more care taken in its engineering and capture than studio PCM for popular music. Upsampling Redbook PCM to DSD is objectively worse and silly. Invest in better reconstruction filters or accept the source was a poor recording.
I upsample pcm to DSD1024 using very powerful modulators and filters from a powerful computer to a 1 bit discrete dsd ladder dac, easily destroy any pcm I've tried. DSDAC1.0 can also do this without a PC is partly why it sounds much better than anything else in its price range.
Jup you don`t need to compress and push to the limiter life as much as the PA is loud enough to kill your ears anyway. But the highest frequency you can reconstruct is at half the sample rate. Infact 44kHz and 96kHz were chosen to get a clean 20kHz or 40kHz signal where the rolloff is neglectable for the human ear.
The real issues are tied to the low pass filters and aliasing which you briefly mentioned. The filters are not there to filter the sample frequency,but the signals above the sample rate. Very steep filters cause their own problems both in recording and reproduction.
Quite. The likes of our esteemed host never seem to grasp the problems with such steep filters so close to the audio band. He (and countless, like-minded reductionists) never stop to wonder if the "crystal clear" sound of digital is actually realistic.
Also the digital synths sounds drastically different on different samplerate Depending on how SRC works in all daws When you render project in different sample rate in Ableton you will hear the difference in mix While in cubase/nuendo you will not
@nicksterj sure, but the quality of over sampling is another part of the reconstruction filter which still needs to be done properly. The same goes for down sampling.
@nicksterj I agree yet some systems still sound better than others. 16 bit 44.1 kHz has enough information if processed correctly for home audio. 20 bit should match pro audio dynamic range. If you have insights on why higher sampling rates often sound better I would be interested. As a note I have worked on DSP systems and have seen examples of poor application of theory so I am not convinced it doesn’t get screwed up in audio from time to time.
"I buy two copies of each CD so I have twice the resolution" ... that had me dying! I genuinely used to be able to hear the difference between 48Khz and 96Khz when I was younger, but nothing more. Specifically when I would have a lot of high pitched distortion in a track with no LP cut off filter. But I'd be pushed to hear the difference now. My ears used to go up to around 22Khz, My left is down around 18-19k these days though and my right a bit lower than that maybe 17-18k (dam DJing headphones). Although sometimes I can tell if there's higher frequencies by how that top end range I can hear sounds, I can't actually hear them >20k frequencies anymore and there would have to be way too much of it for normal listening for it to even be noticeable to me in my audible range. (I've got speakers that go up to 25k, but I can no longer hear super sonic stuff at all sadly, dam age, I just have them so the cut-off is a bit further away from the range I actually can here currently).
The first CD players when they came out in the 80s, had a characterstic “metallic” sound in the high frequencies, because of that steep curve filter needed to eliminate aliasing, which was destroying the phase. Then, overampling came along, generating non existent samples between the actual recorded ones using math, and a much phase friendly 3db/octave for example could be used to eliminate the aliasing, instead of an 18db/octave that was usually needed for the 44.1kHz. The oversampling would make the signal 4, 8 or even 16 times the 44.1kHz original signal, but only to use that kind of filter afterwards. TLDR: Our ears are much more sensible in phase shifts in the higher frequencies than listening to the individual frequencies themselves.
"Our ears are much more sensible in phase shifts in the higher frequencies than listening to the individual frequencies themselves" This aspect is often ignored in those discussions. Consumers do not need higher sample rates for better frequency response, and if that is the thing people focus on when trying if they can hear a difference between 44.1khz and higher sample rates, there shouldn't be much of a difference, if any. But it can make for quite a difference in perception of placement of sounds. Both timing and level differences play a big role in that, and the timing requirements for that are much stricter than one would think when just looking at what frequencies we can hear.
@nicksterj An article with a fair bit of hand waving, which requires very carefull reading to get some details. For example, your claim that timing resolution really doesn't depend on sample rate... his article suggests that, but also claims: "Notice that the calculation of the time resolution does not include the sample rate. Nevertheless, it can make a difference due to the frequency of the signal being part of the formula. A higher sample rate permits higher-frequency signals, which means smaller time shifts can be measured at a given sample resolution" That is simply deception, and while glossed over in the remainder of his article, it contradicts the exact conclusion you took from his article. Additionally, even he is honest enough to point out how his timing resolution claim is very much affected by the choice of the signal he uses, and how a still unrealistic input which is a bit closer to what we'd typically see in music, will cause an order of a magnitude bigger time error. When doing a quick read, his article is highly deceptive, and you for example have been mislead into believing sample rate has no relation to timing resolution, while a much closer reading tells you it totally does, even according to the article you reference. The real points he makes is that it is too simple to equate sample rate and timing resolution, and that bit depth also affects timing resolution, and that is not wrong as such. This is actually well known, but neither of those result in the conclusion that sample rate and timing resolution are not related, merely that the relation is more complex than a simple 'sample rate equals timing resolution'.
@nicksterj " But as I understand it, if you shift the phase of the input signal by some fraction of a degree, the sample values will change by a certain amount - at some point you reach a small enough shift that all the samples will be quantized the same way as before, putting a limit on the difference in timing that you can resolve." That is correct, and explains why bit depth also affects timing resolution and the relation between sample rate and time resolution is not at all the same as the interval between 2 samples. But it does not mean timing resolution doesn't also depend on sample rate. Rather, both affect resolution, and not just timing resolution. This is evident from very high sample rate 1 bit streams like DSD, which have very low sample width, yet produce excellent resolution both in 'levels' and 'timing'. As a general rule, you can reduce sample width as long as you increase your sample rate enough, and the other way around, you can reduce your sample rate, as long as you increase sample width, but within reason. Your sample rate must at the very least be high enough to cover the frequency range you want to capture (ie, 2x the highest frequency you want to deal with).
@@c128stuff I agree with your comments. What I was referring to, considering the first audio CD players, was the analog filters that were used at that time to cut down any frequencies above 20kHz, in order to eliminate aliasing: i.e. two frequencies passing from the exact same samples you have recorded, when you attempt to to play them back. From a pure mathematical point of view, having a window of sound sampled at 44100Hz, can contain a spectrum of frequencies up to 22050Hz if you perform a spectrum analysis, but the same spectrum mirrored after the 22050Hz is also satisfying the same samples. Even though our ears obviously can’t hear those high frequencies, the differences between them create harmonics in the audible spectrum, and that’s why you want to cut them off. So if you wanted to have a frequency response from 20Hz to 20kHz from that sample, you’d need an analog filter which is flat in that frequency range, and then drops dramatically from 20,000Hz to 22,050Hz. That was the one introducing that phase shift I was referring to back then, considering the analog electronics of the ‘80s too.
So two ends of the range are Niquist sampling with infinite bits per sample giving infinitely accurate quantization , and on the other end is infinite sampling rate with 1 bit per sample. And we choose to work somewhere in between, giving best quality / size tradeoff
The thing that is nice with higher sampling rates is that you can line in feed directly a very good 44,1 khz recording and upsample a higher signal thru a ok Dac. Then pass the output to a console and finally to the recording input of your sound card. The console is really interesting because here you can modify the sound and induce a surround effect to the dac output signal and makes it actually produce audio to those frequencies above 22khz all the way to 48khz, within a 96 khz recording sample.
I have digital samples of each pipes sounds of my vitual pipe organ recorded at 96 Khz/24bits. Up sampling from 44.1khz won't give the same detailed timbres as shown by waveforms. When recording with High quality sound card, the ADC is in hardware so it won't affect much CPU cycles. OP says he can't hear the difference. But OLDER people like him has "presbycusis". That's why opera is mostly attended by older people since young people finds opera irritating.
At 192Khz reel2reel archiving, it also captures the 50-70Khz BIAS oscillator faintly from the original recorder. Now this oscillator drifts as the recorder warmed up (tube era), either its mass, or surrounding environment. If it was stable, it could have been a better source to calibrate the motor/mechanical speed shifts shift on original deteriorated tape, but many times there is faint 50hz hum, assuming that was more accurate, I use that.
Virtually all ADCs of the 90s were 1-bit 64fs. In other words, DSD is just recording the first stage of converting to PCM and does away with the need for brick wall digital anti-aliasing filters, as well as reconstruction filters on play back. Nowadays, most quality converters are 2-bit or even 3-bit, at 128-256fs but the real world performance isn't that much greater than can be achieved with 1-bit 64fs ADCs. The main difference is you can get away with not having an analogue anti-aliasing filter (usually 5-pole) in front of the ADC.
Wauw, there is some deafening truth on this channel. Love it! As I remember being told back then, that the record companies where affraid for their buisness when the DAT came out, since you could make a digital clone of a CD, hence the 48kHz. For a long time it was impossible to make a digital copy to DAT from a CD-audio, or record analog to a DAT on 44.1 kHz. Most of the video-broaadcast is still working on 48 kHz, and they won’t upgrade to 96kHz anny time soon, since the benefit of audioquality is minime compared to cost of doubling the data rate. But for acoustic music production I would strongly advise to record in 96 kHz 24 bits, mix in the DAW in 32 bit flow or better, and only SampleRateConvert after mixing. SRC can induce very nasty audible side-effects. Imho the benefits for recording and mixing at 48 kHz are lost with SRC if you need to go to 44.1 kHz for CD-audio. Our ears can only hear maybe up to 15 kHz or so, but what about the harmonics created by insruments above 20 kHz? Take two oscilators, one at 25 kHz and the other at 30 kHz. Now vary one of the two, and listnen to the harmonics you can perceive with your ears. I presume most fo these harmonics are lost when recording at 44.1 or 48 kHz.
24bit 48Khz for recording, mixing, mastering (24bit 96khz only for raw recordings and source archive purposes; for special applications, scientific, we might go for 32bit 96kHz but those are not really for music) If we have 24b\48kHz, then DSP (digital sound processing) should include 'over-sampling' (×2 or ×4 the orig. freq. 48kHz in our case). Record near 'hot' levels (test with low + percussive sound for worst case scenario of a small 'headroom VU' - eaten 'volume units' by the low freq. test sound, and louder Peaks - the percussive hit), if clipping occurs for a few samples DO NOT overthink it - they can be restored in post-recording\pre-mixing production!
Great video and I agree with mostly everything you said. Whenever I multi-track, I use 24/48 as it's senseless to use anything else for the reasons you mentioned. When recording my vinyl records however, I always use 24/96 as that is usually what the master used in the process was given to the cutting engineer, on recent releases anyway. Anything vintage on vinyl I just use 24/48, but new releases on vinyl are truly magical and I don't want to miss a single harmonic.
"I always use 24/96 as that is usually what the master used in the process was given to the cutting engineer" No. normal Vinyls were not even close to that. Commonly used cutoffs were in the range of 17-19kHz to not stress the cutter unnecessarily, with dynamic range rarely touching 16bit. Recording with higher fidelity for preservation is fine, but don't kidd your self into thinking that vinyls are even close to being capable of reproducing those numbers.
I think they meant the masters that were given to the mastering engineer for vinyl were 24/96, not the final output. But to take away from your point, but interesting to see that the final output is so more constrained due to vinyl, crazy! But vinyl can still sound amazing.
My late father (who did a lot of voice-over work) was acquainted with a prominent sound engineer (in Australia) who did a lot of work recording/engineering orchestral music (classical and sound track) who 'bounced' multi-tracks to stereo _through an analogue desk_ (back to digital) because he found the results to sound _considerably_ better than using a DAW's 'sum and differencing'. A digital zealot and reductionist like our host will, of course, find the idea preposterous. BTW, 2822.2 kHz is the sampling rate for 'DSD64' as it first appeared nearly 30 years ago (as an industry archival format, not an "audiophile" one - most people seem to be ignorant of this) higher frequency/resolution DSD128, DSD256 (and even higher still) have long since been available (thanks to those nutty audiophiles, presumably).
As a producer I work with a lot of legacy 44.1kHz 16 bit source material (CDs). My DAW is set to work with a resolution of 24bits, but I tend to stick with a project sampling rate of 44.1 kHz, beholden to the idea that converting from 44.1 to 48 might create artifacts in the audio, and not changing the sampling rate is best practice. I suspect I might be dwelling in the stone age here, but I'd like to try and catch up!
We should consider this debate over, and start discussing lossy/lossless compression formats more! Audiophiles are saying it's all "mp3" but modern compression can achieve way better sound while using less bandwidth/storage. The Opus compression kinda blew my mind, as it is maybe 3x more efficient than mp3 (and used by youtube for its audio).
Great video, it's drove me 30 years back, when I was a student in university. I clearly remember "twice-samplerate-difference", Kotelnikov theorem. All the rest is much simpler: more bits (16/24) means more accuracy for DAC, 48kHz is more often than 44.1 :) Just subscribed, and thanks for your job!
More bits DOESN'T mean more accuracy as long as proper dithering is used. It means lower noise floor. Consumer audio needs about 13 bits of dynamic range for the noise floor to be inaudible in all reasonable listening scenarios. Clearly 16 bit is enough in consumer audio while 24 bit is ridiculous overkill.
I don't care about bits or KHz or sampling rates. All I know is that digital music sounds so much better than records and tape. While I still love my retro component audio system from the 80s, my "go to" listening experience is my tiny Sony MP3 player and a pair of Logitech THX speakers and subwoofer or a good pair of Sennheiser headphones. No hiss, no rumble, no clicks or pops, perfect channel separation, bass I can feel even at low volumes, highs so crisp I swear they could break glass.
Mp3 is mainly sacrifying high frequencies (to obtain its low bitrate of max 320kbps). With 320kbps you can loose 60% of the information in the audiofile..... that s a lot and you would hear that on any decent audioset (not a tiny mpr3player).
@@thepuma2012 i think he wants to say that even an mp3 sounds better than any old audio medium from the 80`s which i agree with tbh. I mean there are some good Vinyl Decks but they dont sound „better“ they just sound „different“ imho and i have a decent Deck and enjoy Vinyl once in a while but if you compare for example bass heavy, (relatively) modern music on Vinyl and the same Song-Track on mp3 320 its day and night. You cant push Vinyl too much when it comes to stereo information. You need to stay much narrow-er in the lows and in the highs on productions meant to be cut on vinyl. Vinyl simply cant keep up on a technical level.. But i think it`s predominantly a matter of taste. Tape also sounds cool and i basically grew up with vinyl and tape but technology has simply moved on. I had the best and most expensive sony walkman (the ones covered in chrome) in the 90`s and then Minidisc came and i had a friend saying „The worst Minidisc walkman is tenfold better than any tape walkman“ and at first i did not want to believe him until i compared them. It was simply day and night To your "60% loss of information“ i can say that i have encountered very very few people who can actually hear a difference of 320kbit to Cd resolution given that you will use a good D/A converter for both. My cousin for example will throw all my arguments and say vinyl sounds better. Its a funny subject thats for sure
@@saardean4481 ok. I understand all you say about the vinyl and 80s audio. But not hearing mp3 to CD difference with a good DAC i find surprising. Indeed a funny subject as you say 🙂
@@thepuma2012 Surprises are the salt of life. I am worried more about the increasing lack of dynamic range and abundance of distortion in modern music rather than mp3 artifacts. As for your reference " But not hearing mp3 to CD difference“ maybe it would help to read again. I said " i have encountered very very few people who can actually hear a difference of 320kbit to Cd resolution“. If you on the other hand have encountered many people that can easily distinguish quality differences between 320 and cd in a blind test then you are a lucky person surrounded by gifted people. You should treasure this 😉
@@saardean4481 Absolutely agree about that "loudness war" issue! I do have a Blu-ray player, but i refuse to buy any audio blu-ray because all of them have compressed DR - even so that an LP version of the same music can have more dynamic range than the disc version. Also on stream-media that problem.
I believe that some people used the 48khz rate because that is what film and television production requires. It fit into their standard of frame rates. SMPTE frame rates, etc.
For me, it is probably the quality of the sound engineering. If it is recorded in 24 bit 44 or above, it means they more than likely care and it has gone through a more modern process and mastering. The difference in mastering quality is massive between, say, an 80s song and a song in 2020s. A lot of the cd era 90s pop was mastered so poorly. The cds sound metallic and harsh, unlike the older tape process. The tape process was warm but compressed. As an audiophile, it is literary hard to listen to older cd mastered music anymore, most likely due to the bad mastering processes used. If you find them in 24 stream or flac and it is remastered, it sounds very different, not because of the bits but the process of mastering with modern day tools. Would love to see a video on this... ❤️
I chose 48kHz, for technical reasons related to streaming. I can run run all devices at 48kHz, but not all of them at 44.1kHz. And dealing with multiple sampling rates can be a pain in delays caused by conversion. So while 44.1KHz is enough on the audio side of things, 48KHz is good on both, audio and technical handling.
As for the "utility" of it, well to each his/her own. I certainly see a use for a medium capable of storing more information than apparently necessary which is down to documentation/curating purposes.
If you record and mix in 44.1 or 48, then I suppose that's typically OK provided that the digital effects are oversampling to remove any aliasing from the application of those effects?
What makes the most audible difference is the objective distortion at low-but-perceptual levels (down to, around, -60dBFS). This is why multi-path architecture excels over any legacy single-path design.
Gotta love audiophile Phil. I feel as if I've known him all my life! I know he is just - as it were - your alter ego. So every time I see him, In my head he's audiophile Shill.
I don't think the question is what sampling rate is overtly better but how the audio is interpreted and reproduced. With a lower sampling rate information about the shape of higher frequency wave forms is lost. For instance consider a square wave at 16khz. At the simplest a square wave requires 4 points. If you sample at 32khz that leaves only 2 points to depict a square. This is a problem even at higher sampling rates (up to 64khz for our 16khz square example), due to the waveform being represented by too few points. This translates into different frequencies in the Fourier transform that make up the signal, which change the harmonics and how the tone is perceived. It's possible to argue the limitations of a speaker system and general consumer use. However, to ensure the original intentions of the sound are most accurately portraied the highest fidelity source should be used.
@nicksterj On most systems sure, speakers will dither the waveform. But the idea isn't what your speakers can produce, but the fidelity of the signal you provide them. Work backward from 44khz, the best possible square signal is only 11khz. And for a 32khz sample rate, 8khz. Now also consider how the fidelity of any audio signal can be expressed as the difference between the sum of infinite sinusodal frequencies and what the source is capable of reproducing. In any case lower sampling rates always reduce the ability to reproduce a signal.
The critical part in the whole chain from analog sound to samples on a cd is the anti alias filter. What is the cut off frequency, how steep is it and what is its phase spectrum. This is an analog filter, applied before sampling. So how does this work in practice, that is what I like to know.
Considering so many speakers use metal dome tweeters which break up with a huge peaks (probably 10 dB or more) above 20 kHz I happen to be believe 44.1 kHz most be best. Not that I believe I or most others can hear or sense it. But why give the dog or local bat a migraine attack? We must be nice to the animals 😊 While writing this it came to my attention that while bats can hear frequencies up to 110.000 Hz the Beluga Whale can hear up 123.000 Hz. It must be quite a challenge for our whale audiophile friends to find underwater speakers to serve their needs 🤔 Greetings from Denmark
Since modern AD and DA converters work with pulse density (1 bit), the precision must decrease with increasing sampling frequency while the chip master clock remains the same. However, at 96 KHz you have twice as many samples with less precision and an anti-aliasing filter with less phase rotation and better transient response. I would consider 96 KHz to be the best compromise for real analog sources, although for my ears it ends at 18600 Hz. I've never been to a disco.😉
24 vs. 16 bps guarantees 2-3 times as big FLAC files. What I often do is bringing the bits per sample down to 20-22 (so there's zeroing the LSB, technically still 24 bps), via dithering. So a lot of my audio ends up as 48/20. Storage is cheap, sure, but with high sample rates and high bps space goes out of hand. Converting FLAC source to target devices, I always go for 48/24 (even upsample 44.1/16 with SoX), because most devices handle it natively - no resampling on the fly.
i don't wanna get into all this samplerate talk rn but 32bit is surely useful in production, cause the resulting wav files don't clip at 0db, so they are easier to work with than 16 or 24bit wavs. if 16bit wavs had that feature I'd likely produce at 16bit because that's what my soundcard supports
32-bit float would have been a bit off topic for this video, but for files created during production, perhaps to share with a collaborator, then yes it's the way to go.
I had trouble putting my finger on what it is about Phil that I like so much but I think it's the confidence he exudes while he speaks his over the top audio-science jargon. He's like a skillful politician who makes you want to believe that what he's saying is true.
Greetings David! What you propose is always very interesting and coherent 👍👍👍👏👏👏 I am an audiophile and an old professional speaker who has been through many recording studios and I have directed radio and audiovisual post production, starting me in the era of open tapes for professional recording. I don't have your professional knowledge but I have always been interested in the subject. I have read some articles and watched some videos, which comment on the harmonics of each fundamental. They mention that within the audible frequencies of 20Hz to 20 KHz, there are harmonics at different frequencies, which can even reach from 49 to 225 KHz. In one of those videos they showed comparative measurements between CD reproductions and streaming, noting in the latter provided by the web, the lack of harmonics in most of the known platforms and only a couple of them reached almost 50KHz in their harmonics. while the CD did it up to 225 KHz. Of course, on the other hand, there is also the issue of the quality of the original recording, as you always refer to. But obviously, even if young people do not hear frequencies higher than 20KHz, an amplifier and speakers that have a frequency response much higher than the audible one would, in theory, be very beneficial for those harmonics that end up configuring each fundamental wave. What's your opinion about it? Greetings and thanks for the topics you present.
I think it's useful to put a figure on frequency response and if there are humans, no matter how young, who can hear 20 kHz, clearly 20 kHz is only just enough. I'd probably double that to 40 kHz, as in the spec for hi-res analogue. Any more than that is a) definitely wasted, and b) asking for trouble.
Is it not: For finished audio files, just to the listeners the resolution makes more difference than the sampling rate when the master properly is converted to 44.1. But for music production and recording the 96 kHz is required to minimize the LPF filter distorsion and rounding and all of that since the audio will be processed and maybe converted more than one time
48K for tracking - as my channel expanders are ADAT, I'd lose half the outputs on each 8 channels at a higher rate, but 48 still sounds great for me. One benefit - as I understand it - with higher sample rate capture above 48, is the files work better with elastic audio processing ITB.
Probably the later. But to the serious stuff. I don't really know if I am correct, but Hemholtz theorized a bandwith--acoustic and non-recorded of course, of around 46khz, because from his experiments he felt that young teenagers could hear up to 23khz, or it might have been 22.5khz. So he wanted a frequency rage twice that-- 46khz-- so that all harmonics below 22.5khz would be able to interact with the information above it and produce combination harmonics in the audible band. Say 20khz combined with 25khz would produce a combination harmonic of 5 kHz, though of course at alower level than the original two tones. So you need to get to 46khz, so that 46 and 22.5--the highest audible frequency--would produce a combination or difference tone of 23.5khz, which would be out of the audible range of hearing for a young teenager with no hearing damage. That was his theory and was why he wanted a bandwidth of 46khz or so. It is also why some manufacturers in the past, especially Harman/Kardon, and I'm sure some British brands, emphasized large bandwidth amplification and the good reproduction of square waves, which is an indication of bandwidth. Essentially, they wanted flat response up to 46khz or beyond. The great Stuart Hegeman engineered some design that went up to 300khz. And the square squarewaves measure by hi-fi magazines showed that. This was in the 60s and 70s. Harman/Kardon still designs this way even though the Harman group, which included a large professional and recording division, and notably, AKG of Austria, was purchased by Samsung, the giant South Korean manufacturer, about 8 years ago. Sad but true. But getting back to the central point. I know I can't hear that high, and I'm not really sure if this matters or if Hemholtz was correct. But the issue is interesting, or intriguing, as the British would say. Maybe we do need a bandwidth and a sampling rate of 88.2 to get all those harmonics into the audible range. I really don't know. As for bit depth, "whew, that's a complicated matter". I guess it can get you lower noise and a larger dynamic range. I can't say if it is of practical use.
You need high sampling rates to reproduce transients as these contain multiple harmonics of the original frequency. The harmonics in isolation are above our range of hearing but the effect they have on the shape of the waveform is detectable as a very sudden transient. As for bit depth it is just not about dynamic range but resolving power and again to reproduce music which is not a smooth sine wave but a messy waveform loaded with harmonic variations then 24bit does a better job.
Yes this is a thing for music creation. As well as low bit-rate, I rather liked the aliasing in the Sequential Circuits Prophet 2000 and was rather disappointed my Akai S1100 didn't have it.
Interesting video. Thanks for that. Making a jump to the real world. Our hearing is analog and we use digital as a bridge. I went to a Roger waters where he performed " the wall" and during the act they build up a wall. at one moment you don't see the band anymore and you are looking at that wall. I made the joke that the band is likely backstage having a drink while we now listen to a tape. But that's in essence the best thing that can happen. If I can't hear the difference between a live recording and the playback I'm 100% satisfied for I'm there when I close my eyes. And what comes closest to that ? is it still tape ? is DSD the best thing ? or PCM ? I don't know for I don't have that reference.
Thanks for this - can you do a segment on this for bluetooth codecs? They're all over the place with aptX lossless, LDAC (which seems to not actually ever achieve its highest bitrate due to RF interference), LHDC (which no device supports), AAC (black box), etc - there's no clear & intelligent source of truth on this subject
Bluetooth is a topic I've been putting off for a long time. I might put it off a while longer. I promise I'll cover it when I reach 100k subscribers. (That will give me some breathing space)
I used to mix in 48Khz, while this maybe enough to convey the mix to any listener, the math behind it or rather said the simulation of analog gear will be less acurate compared to 96Khz And for those who work like people in the analog domain, comiting to a sound or dymamic feel is mandatory to keep the computer fast. I always commit virtual instruments with loads of plugins and bounce them to an audio track. If I could i would rather mix on 192Khz that is what most commerial studios do and they are not wrong. Mr. Neve has said this and he is right, there is an psychoacoustic ellement in the high frequencies that is experienced by listeners, this was one of the revelations mr. Neve had during his life.
This is new to me that the higher rates are purely for capturing higher frequencies. What about the higher capturing fidelity of the waveforms? That's always a good thing if the pure signal dictates the sound more, and not the DAC. It isn't very critical of a topic looking at pure single frequencies, high or low, but in the case of a complex sound, the higher sampling rate, the less it all becomes white noise in the end.
Recording is done at the Nyquist rate so double the sample rate. e.g. 44.1 recordings are sampling at 88.2. Doing so already preserves 'all' the signal you input upto the sample rate. Nobody can hear above a 44.1 sample rate so recording information above it with even higher sample rates is pointless, unless your analogue converters do actually sound better at higher sample perhaps and/or the conversion to lower sample rates sound better than just recording at them. e.g. a drum being pitched up or down will pitch down more cleanly if you recorded at higher sample rate etc.
In my experience 96k gives less sibilance, less need for de essing and corrective high end EQ, better sounding pitch shift and warping. The best place to listen is in a vocal or a snare. If I can’t hear it immediately, after processing and limiting and on multiple tracks it becomes night and day. If I need to save space with other tracks il make sure the vocal and snare at least are 96
The reason I heard for the 44.1Khz was because the human hearing has a frequency range of 20Hz to 20KHz and, according to Nyquist, to convert an analog signal to digital without any loss you have to sample it at twice the rate of the maximum frequency. By sampling it at higher than 44.1KHz you're just losing storage space.
Yeah, its due to that as well but the exact sampling frequency apparently originates to that audio recorded to a video recorder that he spoke about, i havent heard about that before this but it was common earlier on to use the same formats for better compatibility betweens systems, nowdays things are often far more flexible. I agree, 44.1KHz is already "full res" anything over that is mostly wasted space and CPU power, i record my band in 44.1KHz 24 bit in my DAW for our albums.
Well, there are benefits of working with higher frequencies and higher bitrates for a DAW internal audio engine, allowing more accurate multiple tracks mixing and audio processing. Obviously afterwards, for whatever actual output format, 44.1/48 kHz are good enough. Working with higher bitrates also allows a lower input latency while still having a decent/large buffer size (avoiding buffer under runs). With the newer CPUs, I feel 96 kHz or even 192 kHz shouldn't be an issue but unfortunately there are plug-ins (VST64/VST3) which won't work if your DAW is set to such high sampling frequencies... I found myself having to limit my expectations and stick to 48 kHz with a pretty high buffer size (2048 bytes) if I want to be able to stack many synthesizers with no crackling... The input latency is pretty bad but I can live with it.
I read not too long ago in some forum that 96 kHz sampling is the thing if you are thinking on releasing vinyl records (analog). I don't know if there's any truth in that statement or it's just BS as per usual. Discussing 32 bit float soon? Just subscribed to your channel. Thanks for the video. Cheers.
I recently purchased the S.M.S.L. su-9 pro, because I wanted to be able to decode DSD to DSD512, and 24bit/192KHz for playback with my nearfield listening system. Trying not to be the audiophile that Alan Parsons refers to as listening to the equipment, and not the music, the main difference I think I hear, is not audio quality but time correction, and that probably has nothing to do with the sampling rate at all. My playback is though a 1975 Pioneer SX434 into Wharfdale Diamond 9.1's which probably produces the quality of sound somewhat equivalent to my aged hearing
High sampling rates for playback are insane 48K is the accepted norm for all platforms and is a happy medium but even 44.1 is fine for playback in the final product. But for music production using high sampling rates like 96K is a bulwark against aliasing distortions. Its not about the bandwidth and how much of that which you can hear. But high sampling rates such as 96K pushing aliasing distortion out of the hearing range that would otherwise creep down to the audible range at 441 or 48K. It helps when using digital processing like saturation, and anything that adds harmonic content to the signal, it also gives some EQs the ability to not cramp. I heard a professor from ETSU's music school describe sampling rates in a very interesting way once. I wish I could animate it because its a great visual. He said imagine standing in front of a 10,000 feet tall concrete wall with a concrete floor and you are standing there with a tennis ball (aliasing distortion) in your hand. Imagine standing 21.5 feet away (44.1K) from the wall and throw the tennis ball as hard as you can and where ever it lands that is where aliasing would be heard in the hearing range. Now move back to 48 feet away (96K) and perform the same throw. Most likely at 21.5 feet the ball went over your head and landed behind you somewhere, but at 48 feet the ball is less likely to even make it back to the spot where you are standing. I thought this was a really great visual for describing aliasing.
@@RockManEnough Oversampling in converters can help give the filter a more rounded shape, but the fact remains there is more to a session than just recording the audio through the converter. the antialiasing filter is still in the same place. So no matter how many times the chip will oversample x128 or x512 the filter at 48K will still be at 24 kilohertz on the output leaving the antialiasing filter in the same place same place but with a more favorable shape. Oversampling isn't a replacement for using better LP filter placement. The type of filter used as well is highly important. Lavry explained this in the sampling rate white paper that was published nearly 20 years ago when they were building their next generation conversion chips and they explain why converters with oversampling can mask many of the failures of digital conversion filters. Its one of those deep audio subjects you can't really have a good conversation about in the UA-cam comments section. But saying there shouldn't be aliasing at any sampling rate for converters is true but we don't just record audio and thats it. We also use digital tools and effects that will work better at higher sampling rates. Time stretching, Tuning, Distortion, Saturation and Pitch algorithms all benefit from the highest sampling rate possible because this gives them more data points to work with. But in the end if it sounds good it is good so thats the great part about choice if a person wants to work a lower sampling rate they can make music at 44.1K and if other people want to work at 192K they can. I work at 96K because its the best spot for me but thats the great thing we can work how ever we choose. Cheers, Have a great day!
The guy at PS Audio said on UA-cam that they have started using DXD when making recordings which they then transfer to DSD. That is, significantly higher frequency than 48 kHz, they have a record label called Octave Records Note that guy also answers questions from people on youtube . If I understood correctly, he is the founder and CEO of the company
If you are producing music, the delivery format defines the sampling frequency and bit depth. 24/48k is the norm these days. Multitracking with 16 bits will save on disk space at the expense of a higher noise floor. If you’re applying any non-linear processing then higher samples rate (or oversampling) will help combat any aliasing.
It is absolutely true that 16-bit has a higher noise floor. However it's likely that most or all of the faders will be lower than 0 dB so there's less noise than there would otherwise be. If you're mixing to 24-bit of course.
I very much enjoy your educational videos. Would you consider making a video about harmonics or ultrasonics? Frequencies from instruments higher than humans can hear, that may or may not add to the perceived sound of music when played live or when listening to 96khz or above recordings through resolving audio hardware and or Hi Res headphones. I keep reading and hearing of this perceived sound that subconsciously makes the sound we hear more real. High frequencies that our body can sense but not necessarily hear, like sub bass below 20hz. I'm wondering about the benefit of adding a pair of super tweeters to my home hi-fi system. Would love to learn your thoughts. Zach
I have recorded with 24 bits at 96Khz for years, not because I can hear a difference in sound, but because I believe that I get greater headroom and much less aliasing with this choice. And as I typically work with about 100 tracks in a song, reducing the problems created by lack of headroom and distortion from aliasing seemed like a good idea, even if there seem small. It costs more to be able to do this, but I do it anyway. Maybe I am wasting money and effort ?? Thanks!
Thank you for this topic. 1) Sampling rate for Capture: I do many conversions and audio restoration from analogue LP and RTR masters to digital for the purpose of remastering. Spectrum analyzing the musical signals of an typical recording reaches high frequency of around 40Khz and better quality reaches even up to 60Khz (!) If I want to fully capture the whole information I need to use a sampling rate of 192KHz. However, the 192KHz is a difficult task to handle in many aspects' including the A/D performance (Signal/noise) and from practical reasons I use only 96KHz/24 Bit Sampling rate for the capture and for the post processing. 2) 20KHz and above musical information: The signal level may be low at these frequency but the energy (Power x Time) is high. if we would like to fully preserve the full musical information we must sample the signal at high sampling rage. 3) Can we hear the Signal above 20KHz? We can't Hear or identify tones (Frequencies) above 20Khz, but we are capable of feeling the presence of high energy of high frequency when it's envelope is modulated by an audible signals (similar to the principle of how we can recognize a dithered low level audio signal). 4) To my experience the 96Khz sampling rate is a must from capture till reproduction. 5) D/A Filter Slope: When sampling a 20KHz audio signal at 100KHz there is an aliasing effect around the 100KHz from 80 to 120Khz where the 80Khz artiact is generated by the 20KHz signal and 99900Hz is generated by the 100Hz signal. When Sampling rate is 48KHz the Alias of 20KHz is generated at 28Khz. There for the LPF of 48KHZ must Pass 20KHz and Must Stop at 28KHz. it is a very challenging slope for the LPF, which presents heavy phase distortion within the audible range. 6) 1 bit: provides a very accurate linearity for low level signals capture and reproduction. has lower bit resolution at higher frequencies as well as higher noise level than a PCM. Thanks Again.
@@NamelessSmile a high frequency can swamp a gain stage -- you can't hear the frequency that's causing the distortion, but you can hear what it does in the audible range.
of course you can get high frequencies off of tape: the bias signals for most decks are up in the 30kHz+ range, and if you can recover them, you can correct for wow/flutter. (to good effect, as well!)
Anything over 40khz enables one to perfectly reconstruct signals below 20kHz. The problem is impact type sounds ( like percussion) , for one, exceed that 20khz bandwidth and even though you cant hear a 20khz sine wave, you can hear the sharp rising waveform of a drum thump as well as “distortion” on real instruments. Then also, the way the signal is reconstructed- rarely do the compute inverse laplace transforms etc - they just interpolate. Hence the divergence of opinions
Distortion and wild frequencies make their way up past 20khz into ultrasonic ranges. You can get mics like the Sennheiser MKH8000 series that can pick up 50khz. Anything above 20khz isn’t just noise. There are harmonics up there just like there are resonances all over. Sure, it’s not too practical. Not a lot of people recording bat sonar pulses and slowing them down, but they do exist. 88.2 is probably ideal. You want to be able to capture at least upper harmonica and resonances that make distortion sound brillaint. 192khz and 176.4 are great for sound design. You have to think about if you might need to pitch shift something. If you’re just going to be recording nursery rhymes and polka, you’re absolutely safe at 44.1 and 48 DSD goes at some crazy high rates 5.6mhz and 11.2. Might even go up to 22.4. I can’t remember. And yes it sounds amazing.
@nicksterj I have qualifiers in there like pitchshifting, and "Sure, it’s not too practical. Not a lot of people recording bat sonar pulses and slowing them down, but they do exist." What you're saying is practical, but there are esoteric reasons beyond, which makes it worth it. Especially archival purposes. It's like saying "my jpg or bmp files are perfectly fine because they have all of the colors I like in it already. I'll be deleting my raw files" No, the raw files have extended data that you can access and remaster later. Try recording a lot of distortion. I can hear a distinct difference on my system. At 44.1 and 48khz it sounds rolled off. Not just guitar distortion but going through something like a Thermionic Culture Vulture or something that can introduce a lot of pentode/triode distortion into a layer. It gets tricky at high frequencies you're dealing with a lot of micro distortions essentially giving you high freq square waves. They'll capture at 44.1/48, but the math gets a little fuzzier. Also like the guy said in this video, at higher frequencies you get a better rolloff. I think going too high in frequency like 192khz, unless you're doing sound design, is a bit overkill. The sweet spot in my experience (probably going on 27, 28 years) 88.2/96 is the sweet spot. I can see the arguments for going 44.1 to save on processing power, but computers can handle it these days.
New advances in computing are looking at replacing digital processing with analogue processing -or digital analogue hybrids- due to being more energy efficient, etc. Well sound processing and storage may go back to analogue again?
I love your sense of humor. You have a very clear understanding of the tech & you methodically dismantle the BS that so many pseudo-audio-experts spew. I record in 48k/24bits/chan, operate filters in 8x over-sample mode to minimize aliasing induced distortion. Final mix down to shaped-dither 48 or 44.1 in 16bits.
I only like listening to the highest level of audiophile technical cheekiness and this channel is cracking! Another well done episode mate!
i think you didn´t understand the question
Finally another fantastic video about this topic besides Dan Warroll’s. We don’t think about needing to hear 96,000Hz frequencies when recording at 192k. What we are thinking about (in terms of auditory effect) is the blending of digital audio samples as they combine together in a virtual space. This “smoothing” effect becomes much more apparent on things like cymbals in drum overheads & instruments with longer envelopes that ring out.
When combined with digital (or virtual) signal processing like high shelves, you are able to use more EQ with less apparent artifacting and aliasing when compared to lower resolutions. Some plugins compensate for this by upsampling inside the plugin for processing, then downsampling on the way out, but not all of them do, and I haven’t heard of a DAW doing this with their console strip (please correct if I’m mistaken).
The argument listed here about CPU power becomes a bit more null & void as time goes on. 192kHz is also much better for any kind of time or pitch correction, since higher resolution gives more samples to stretch & blend. The last benefit I’d mention for 192kHz is the lower latency times on system buffers. On certain systems, this is extremely desirable and beneficial since desktop computer systems these days can handle the load with much greater ease than ever before.
Comparing me with the legendary Dan Worrall is almost as good as being awarded the coveted UA-cam play button (which he has and I do not).
This seems to come from the "stairstep" myth of representing samples. Sampling theorem states that everything within the bandlimited signal is captured . Thus, 192k does not capture anything within the audible range any better than 48k. This is easily provable with analog equipment as was demonstrated in this video ua-cam.com/video/cIQ9IXSUzuM/v-deo.htmlsi=KF5VEVHFq9ZM8885
"This “smoothing” effect becomes much more apparent on things like cymbals in drum overheads & instruments with longer envelopes that ring out."
Try it - but before you get disappointed: The most well payed audio-engineers whose sole job it is to validate audio quality can not head a difference where there is none.
@@ABaumstumpf I’ve gone back & forth a million times on it for the past 20 years and found slight, but very noticeable differences in such cases. Not everything requires such high sample rates, but it can transform how all of the tracks work together on a mix, and ESPECIALLY when hard quantizing drums.
Sounds more like you should try it, many, many times before relying on what someone else says about it.
@@JesterMasque Higher sample rates are massively important when doing any kind of intence time stretching in the digital domain. They can also help stop or reduce aliasing when you don't have oversampling avalible natively in a plug in that would benafit from it or in your DAW. If a top engineer can't hear aliasing in the audiable range, I'd question their position, and by using souly lower sample rates you will surely run into it if using diferant plug ins all the time.
In terms mastering from tape to CD, what makes most difference is: (1) the quality of the source tape (2) subtle changes to eq etc during transfer to bring "the breath of life" to the presentation (3) the quality of analogue to digital conversion (ADC). The limitations of early ADCs was recognized by Tony Faulkner, who modified Sonys for better performance & the team at Pacific Microsonics who were developing HDCD.
Was entertaining to listen to and glad that it’s pretty much factually accurate up till the SACD as really you missed how the supper high frequency low but rate encoding actually works and why the recording industry would want to use this for archival and data storage. Well worth listening and even subscribing to. 👍🏻
48k/24bits I looked at the options and those were the numbers that made most sense to me. Everything sounds great, I have flexibility.
As an electronics engineer, I really enjoy our down to earth technical videos, keep them coming! Regarding DSD & SACD, it's no coincidence that Mobile Fidelity Sound Lab use it for their digital releases... From a hardware engineering perspective, a bit stream D to A is much simpler than the alternative so it's much easier to do it right.
How do you mix the DSD streams? Or do you just release the recording without any mixing?
@@ianhaylock7409 conversion between bitstream and PCM is trivial and transparent.
@@ianhaylock7409You have to convert to PCM to edit/blend/EQ/whatever, then convert back to DSD
Quality, you do make my day!! Thx for sharing🎉😂😂
3i just subscribed. You are a fine young man who seems wiser than their age.
Actually I'm hoping for my wisdom to catch up with my age. It's taking a long time.
I just love how you cleverly are able to push all of the right buttons, without actually having to push them. LOL Another very well-done and informative discussion sir!
Thank you for the video. Have you done a video on how the streaming services manipulate your music as far as loundness and compression when uploading your tracks to their platforms.
I have in mind at some point to compare one of my tracks on Spotify with the original master. Whether any difference will be heard through UA-cam's audio mangler will be an interesting question.
Crazy 🔥🔥thanks for the info
Thanks for this piece, I really enjoyed it.
Having been trained as studio engineer back in the 80's, back when we were getting our heads round digital and just learning to edit without putting some tape on the cutting block, I feel I know many technical matters in regards to sound reproduction and recording techniques, but I learnt several things from watching this.
In addition to my more formal training, I've also been an 'audiophile' since I bought my Roksan Xerxes turntable as a young bushy tailed 20 year old, and have spent the last 40 years enjoying fine home reproduction and keeping abreast of the developments in the industry. 40 years of this has clearly shown me there is still so much we dont know and dont know how to measure. Whilst I'm a scientist by nature and training, Science has always been about explaing the observable to me, not dismissing observations and so I'm happy to acknowledge that amplifiers sound different and that cables sound different. When its as clearly observable as I hear it I accept my experiences, guided by many years of both home and Studio Audio.
I'm almost 60 now and I'd be lucky to hear anything much over 14Khz. But given a known decent digital mastering, I can wholeheartedly tell you I can hear a difference between a 24/96 and 24/192 version of the same master, and its not a small difference. This is not expectation bias as I can get my wife to swap between the recordings (easy with Roon and access to all the files via Qobuz or similar) and identify the recording sampling rate with very high accuracy. My wife just like a good sound, but she has regularly commented on some studio masters sounding astonishing too even though she does not undertsand why they are different. I cant explain why I can hear it so clearly, but I accept it.
It is of course the mastering that makes the most difference to the quality of a recording and I can also point to many remastered Albums in 24/192 or 24/96 which sound much worse than an earlier master on a Japaneses CD at 16/44.1.
With so much confusion and so much we still dont know, its no wander the industry is plauged by those peddling 'snake oil' and their very exsistance gives those of us happy to work with the observable, some resistance from those that confidently state 'if you cant measure it it doesnt exsist'. Its always been a conundrum for those of us who indulge in some audophilia whilst still seeking the scientific answers of how to measure what we can observe.
One subject that some people ask me at times but unfortunately i lack the skill to explain as comprehensively as you is what happens when you
work in one bit/resolution rate and then export at another. Like what happens if you limit a project at 96khz/24bit and export at 44.1/16 and the other way around, what happens with the ISP etc etc. I think a video on the subject would be worth a lot
An correct optimal solution derives the sound wave using interpolation, e.g. cubic spline. Then it resamples the waveform to the new output format.
In a lot of DAWs, at least Adobe Auditoon, if you zoom in to the max you’ll be shown both the samples and the interpolated waveform.
There’s a lot of old misinformation about how very old stuff maybe worked with digital in the 80ies. According to that misinformation there’s a lot loss and strange dithering noise applied to counter act it. My friend tried to explain that voodoo stuff to me and I stared confused at him…
Pretty sure no DAW in modern times uses strange voodoo bullshit. Calculating the waveform and resample the to the output format is the same approach with barely any up/down sampling issues. Why you’d ever go away from the waveform in any application that prioritizes accuracy would be beyond my comprehension.
Maybe some hardware that requires absolute zero latency do weird voodoo still, but DAWs surely must just resample waveform at best accuracy possible.
@@randomgeocacher thank you for the elaborate answer. I’m asking because in the past somebody once told me that for one example, if you use a limiter work in one Bitrate , and then export to another, you might be ending up exporting the file without any limiting applied.I cannot, however remember the exact terminology what betray toward betrayed this applies
@@randomgeocacher A complication with doing things this way is that applying a brick-wall filter to a sampled signal may result in peaks which exceed the peak level of the original signal. As a simple example, if one has a signal which, after sampling at 1 million samples per second, looks like a 1000Hz square wave whose peak values are 0.01dB below saturation, and downsamples to 44.1kHz, each edge will have a certain amount of overshoot before the signal settles toward an equilibrium until the next edge approaches.
Audiophile William here, I've tried and tested many highend DAC's over the years, i never paid any attention to all the numbers just the sound. I borrowed a MSB Premier DAC that does 44.1kHz to 3,072kHz PCM up to 32 bits and from 1 to 8xDSD, in the end i decided to keep my old modified Audio Note DAC from 1998 which sounded just as good if not better than the MSB Premier. My old Audio note dac manual says it has no over sampling, no jitter reduction, no noise shaping and no re-clocking and uses the highest grade AD1865, 18bit stereo converter chip what ever that all means?? Audio Phil tickles me every time 🙂
Thanks for that , Audiophile William !! I don't feel bad now, about really liking my 2004 Apogee Mini DAC ! 😂. I think the clocks and analog sections of DACS have much more to do with sound quality rather than the DACS used .. Almost ALL DACS today are perfectly capable on specs ...
Love the Audio-Note dacs. I have ANK 5.1 Signature dac kit 24/96 nos. best dac I ever heard.
All modern "brand name" DACS like TI (formerly Burr Brown) are absolute overkill for "listening" to CD's. I recently purchased a Rotel RCD1572MKII that uses the TI PCM5252. However it's more than just a $10.00 chip to do a wonderful conversion. The OP amps and output side design and components are also extremely important. In my opinion most US$1,000+ CD players are more than satisfactory to most ears without needing an external DAC. @@LukeSchneiderEWI
the super high oversampling does not make the aliasing issues go away. they have been already made in the ADC when recording the music.
I just listen to music and my latest CD player is even more "sensitive" to crappy recordings/mastering's than anything I've owned before. It is utterly merciless and as analytical as my B&W speakers are in a similar manner. @@stefanweilhartner4415
Even CD quality is "only" 44.1 kHz (16 bit), enabling almost 22kHz bandwidth. That is what 10 year olds can rarely hear, even at higher volumes, and only when not masked by lower frequencies. If you are above 25 years of age - forget everything above 18 kHz.
i am a hearing aid acoustic technician, 27 years old. in blind tests i can just about differentiate silence from 24 khz at 65 db.
its more a feeling than a sound, like how a tense jaw has a "sound", but between your ears and higher.
but at that point its not about any soecific data, but about if this noise has a natural harmonic with the music (not disonant, nit misding, not overbearing).
that does not require actual info on the medium, but simply a harmonic overtone extrapolation with a nice roll off.
so 48 khz recordings, which hits my differentiable 24 khz ears, and 2x supersampling + harmonic distortion ultrasound dsp work that rolls off up there, is undifferentiable to me from 96khz or above recordings.
my ears have a 120db dynamic range between lowest detectable sound and uncomfortable volume, so 24 bit exceeds that range well at 144db dynamic range.
As a kid, I could hear the flyback transformer in old tube TV's. They operate at 30khz. I can still hear above 20khz in my 50's. Not all ears are created equally.
@@Chrisspru you state your ears have 120db dynamic range, as someone who works with ears, you will be well aware of the damage listening to 0db then a 120db peak will do to your ears, you will also be aware of the average DB in a room and how noise floor above this equates to perceivable dynamic range.
The reality is, with a recording, all else equal, no human needs or can benefit from above 16.44.1 no human can perceive the differences, and even funnier is the lack of actual mics recording above 20Khz and the lack of audio equipment to actually have the capability to play a 120db dynamic range, never mind actually excerise the 96db range of 16/44.1
In professional studio work and mastering there is a point to 24 bit.
But in HiFi its pretty much foo like cables to make more money.
@SastusBulbas1 my hearing goes from -10 db to 110 db.
and people can tell the difference between the 16 bit and 24 bit in reproduction just barely with very good ears. a tiny noise difference, an odd feeling that hits 80% on differentiation in blind tests. its pretty weird.
with the hz the difference is in high note accuracy in a similar odd feeling way
@@Chrisspru your hearing is the same standard of hearing every human has.
People cannot tell the difference between 16 bit and 24 bit audio files, this has been done to death by studio engineers.
What people do hear a difference in, is DAC's due to topology, and even differences in mastering.
So far blind tests have shown that audiophiles cant even tell the difference between a cable and wire coat hanger, a Bryston or Alexis amp, and websites these days are full of information as to why this is, Audioholics and Audio Science Review can be quite blunt with measurements and reality vs feelings and imagination.
Always providing top notch content with classy humor
As audiophiles can hear what no mortal who have participated in scientific experiments can hear, perhaps they could donate their ears and brains to scientific research; preferably _ante mortem_ .
I can hear it cos I bought lots of expensive equipment, dacs, oxygen free gold sparkle cables and I have to justify my expense
Brilliant! Ya old bloody bastard! Well done... Ya know it all comes down to a human ear. Just like modern TV's that claim a "billion" colors, when the human eye on average can only see about one million. The same applies here (or should I say "hear"). Keep doing your stuff. Love it...
I think the low noisefloor of 24 bit is convenient for recording but at the same time I think my music does require high sampling rates as well so 24/44.1 serves me well
I’m also too 24/44.1 ✅ is enough. They say higher better. However, it is not always guaranteed that the oversampling results better performance😅😉☺️….
16 bit 44.1kHz is the limit on a CD no matter what you up sample to there is no additional information.
its not about upsampling, but using high-res files from start
What may seem crazy today might just be every day in the future. I’m not sure if this is a version of Moores law but I know my ears are not going to double in frequency response any time soon. Great humour with Audio Phil a supporting cast who knows his thang.
Thank you very much for this highly interesting video!!!
I dithered around so long that I can only offer a truncated opinion that 24/48 is good enough for me.
Thank you for the ever exilarating but serious presentation. I use, if I can choose, 48KHz sampling rate and 32bits floating point resolution (mostly to avoid internal clipping). And I export my masters in 24bit 48KHz AND 16bit 44.1 properly dithered. I have tried 96 KHz and I DON'T like it unless it is for Jazz or Classical recordings. For Pop and Rock I feel is losing some grit (It might be some sort of psich-acoustic impression on me knowing the sampling rate. In general I am fully satisfied with 48KHz and if I can use oversampling in som plugins for extra resolution in some cases I am fine. The most important fact is that if it sounds good (to me) then IT IS good and that's it.. And if in 30 years from now they are still reprinting my recordings, then, good luck. I don't think any sample rate difference will make a difference. But who knows... When I built my first home studio around an 3340S Teac in 1975 I was dreaming to have a way to have a full 24 track studio in my bedroom. And in 1992 I was dreaming about a software I named "de-blender" that was capable of separate the individual instruments from a store recording... And here we are... both are a reality now. How impressive. The next step is a software that mix the tracks with A.I. (hahah... ) But I hope not as good and as artful as myself. Thank you for your contribution to the demistification of sound engineering.
I had an SACD player and the sound stage it could throw out was stunning. No CD I had could do the same. Many years ago when single bit DACs started to hit the market I was on the hunt for a new CD player and it was easy to identify the single bit machines by the scale of the presentation. Don’t really know why but the high bit rate machines just seemed better.
The reason why you never see bitstream used in DAWs is they can't handle the processing that way. They would effectively have to convert it to PCM on the fly, process and then convert back to bitstream.
Excellent info. I always learn something new here. Thank you. 😊
Great video thank you 🙏🏽
What if we took the single-bit sample to an effective infinite sample rate. We could use amplitude to modulate a single stylus moving across a spinning cylinder coated with wax so that the number of samples is limited only by the number of molecules in the wax.
High-frequency response would be limited by the inertia of the stylus. 1 bit sampling at a high frequency is essentially what a class D amplifier does.
The other point about filters that almost got talked about is that steeper (higher order) filters generally introduce unwanted phase shifts. I think a lot of people can notice this.
This is a good point. I mentioned phase very briefly here and it's a topic I will cover in more detail in future.
@@AudioMasterclass YES, YES! You did drop an ph-bomb and I noticed it, thanks!
@@AudioMasterclass As Captain Kirk often said, set phasers to stunning.
Broadcast audio is moving toward 48khz as the recording standard. Used to be 44.1 for a very long time. For playback, it's 48khz, but that is changing. The final link from the digital audio processing to the transmitter exciter is becoming 192khz, so that stereo multiplex and ancillary data can ge generated by the computer doing the processing.
With regards YT's algorithms, I'm not so sure they're clever at all. They show videos on similar topics to the ones that I've watched recently but... they also persist in recommending items about Class 55 Deltic locomotives that my son used to love. Love that is... 10 sodding years ago! BTW... loving your channel :)
Unfortunately the algorithm heard you when you mentioned Class 55 Deltic locomotives and you will be seeing more of them soon. And now so will I.
Higher sampling rates are good for heavy processing in daw. Warping stretching pitching and so on. Especially for slowing down or pitching down.
That makes sense to me :) also if you pair with a specialty small condenser mic that can capture beyond human hearing range (there’s a few that aren’t crazy expensive) you could even recover human inaudible higher frequencies. In theory it could be cool to hear animals like bats well, or recover unheard ranges of chaotic sounds like explosions or crashing glassware..,
Great video. That last bit was brilliant 😂 get it?
Double denim tuxes aren’t for everyone, but certainly brilliant LOL
My sampled instruments are 24/48 so I don't bother using anything other. Equipment is important and ensuring you have good AD-DA conversion and monitoring equipment.
As for venue recording - keep it simple.
Good tear down of the matter.
You are a healthy, intelligent man!
96kHz is very popular in live sound, as it halves signal propagation (processing) delay, which improves fold back to performers.
Regarding DSD & SACD, to my aged, & admittedly rather abused, hearing; I find it sounds far more natural, open & clean (best words I can think of) but that might just be down to how they were recorded & mixed and that they have not suffered from being limited to death as is current CD mixing & mastering practice.
Another DSD/SACD enthusiast like me. I agree with you wholeheartedly regarding the cleanness and openness of the sound. I also find it true with regard to transients such as cymbals and drum thwacks, Keyboard fingering on the piano is more evident too. I have been downloading 5.6448MHz DSF (DSD128) albums from NativeDSD Music for some time now and most of them sound wonderful.
Native DSD has much more care taken in its engineering and capture than studio PCM for popular music. Upsampling Redbook PCM to DSD is objectively worse and silly. Invest in better reconstruction filters or accept the source was a poor recording.
I upsample pcm to DSD1024 using very powerful modulators and filters from a powerful computer to a 1 bit discrete dsd ladder dac, easily destroy any pcm I've tried. DSDAC1.0 can also do this without a PC is partly why it sounds much better than anything else in its price range.
Neither matter. 16 bit/44.1kHz is the maximum that makes sense. Unless you have bat ears and live in a completely isolated room in the countryside.
Jup you don`t need to compress and push to the limiter life as much as the PA is loud enough to kill your ears anyway. But the highest frequency you can reconstruct is at half the sample rate. Infact 44kHz and 96kHz were chosen to get a clean 20kHz or 40kHz signal where the rolloff is neglectable for the human ear.
The real issues are tied to the low pass filters and aliasing which you briefly mentioned. The filters are not there to filter the sample frequency,but the signals above the sample rate. Very steep filters cause their own problems both in recording and reproduction.
Quite. The likes of our esteemed host never seem to grasp the problems with such steep filters so close to the audio band. He (and countless, like-minded reductionists) never stop to wonder if the "crystal clear" sound of digital is actually realistic.
Also the digital synths sounds drastically different on different samplerate
Depending on how SRC works in all daws
When you render project in different sample rate in Ableton you will hear the difference in mix
While in cubase/nuendo you will not
You should not worry about filters at all!
@nicksterj sure, but the quality of over sampling is another part of the reconstruction filter which still needs to be done properly. The same goes for down sampling.
@nicksterj I agree yet some systems still sound better than others. 16 bit 44.1 kHz has enough information if processed correctly for home audio. 20 bit should match pro audio dynamic range. If you have insights on why higher sampling rates often sound better I would be interested. As a note I have worked on DSP systems and have seen examples of poor application of theory so I am not convinced it doesn’t get screwed up in audio from time to time.
"I buy two copies of each CD so I have twice the resolution" ... that had me dying!
I genuinely used to be able to hear the difference between 48Khz and 96Khz when I was younger, but nothing more. Specifically when I would have a lot of high pitched distortion in a track with no LP cut off filter. But I'd be pushed to hear the difference now. My ears used to go up to around 22Khz, My left is down around 18-19k these days though and my right a bit lower than that maybe 17-18k (dam DJing headphones). Although sometimes I can tell if there's higher frequencies by how that top end range I can hear sounds, I can't actually hear them >20k frequencies anymore and there would have to be way too much of it for normal listening for it to even be noticeable to me in my audible range. (I've got speakers that go up to 25k, but I can no longer hear super sonic stuff at all sadly, dam age, I just have them so the cut-off is a bit further away from the range I actually can here currently).
you are full of sheet son
The first CD players when they came out in the 80s, had a characterstic “metallic” sound in the high frequencies, because of that steep curve filter needed to eliminate aliasing, which was destroying the phase. Then, overampling came along, generating non existent samples between the actual recorded ones using math, and a much phase friendly 3db/octave for example could be used to eliminate the aliasing, instead of an 18db/octave that was usually needed for the 44.1kHz. The oversampling would make the signal 4, 8 or even 16 times the 44.1kHz original signal, but only to use that kind of filter afterwards. TLDR: Our ears are much more sensible in phase shifts in the higher frequencies than listening to the individual frequencies themselves.
"Our ears are much more sensible in phase shifts in the higher frequencies than listening to the individual frequencies themselves"
This aspect is often ignored in those discussions. Consumers do not need higher sample rates for better frequency response, and if that is the thing people focus on when trying if they can hear a difference between 44.1khz and higher sample rates, there shouldn't be much of a difference, if any. But it can make for quite a difference in perception of placement of sounds. Both timing and level differences play a big role in that, and the timing requirements for that are much stricter than one would think when just looking at what frequencies we can hear.
@nicksterj An article with a fair bit of hand waving, which requires very carefull reading to get some details.
For example, your claim that timing resolution really doesn't depend on sample rate... his article suggests that, but also claims: "Notice that the calculation of the time resolution does not include the sample rate. Nevertheless, it can make a difference due to the frequency of the signal being part of the formula. A higher sample rate permits higher-frequency signals, which means smaller time shifts can be measured at a given sample resolution"
That is simply deception, and while glossed over in the remainder of his article, it contradicts the exact conclusion you took from his article.
Additionally, even he is honest enough to point out how his timing resolution claim is very much affected by the choice of the signal he uses, and how a still unrealistic input which is a bit closer to what we'd typically see in music, will cause an order of a magnitude bigger time error.
When doing a quick read, his article is highly deceptive, and you for example have been mislead into believing sample rate has no relation to timing resolution, while a much closer reading tells you it totally does, even according to the article you reference.
The real points he makes is that it is too simple to equate sample rate and timing resolution, and that bit depth also affects timing resolution, and that is not wrong as such. This is actually well known, but neither of those result in the conclusion that sample rate and timing resolution are not related, merely that the relation is more complex than a simple 'sample rate equals timing resolution'.
@nicksterj " But as I understand it, if you shift the phase of the input signal by some fraction of a degree, the sample values will change by a certain amount - at some point you reach a small enough shift that all the samples will be quantized the same way as before, putting a limit on the difference in timing that you can resolve."
That is correct, and explains why bit depth also affects timing resolution and the relation between sample rate and time resolution is not at all the same as the interval between 2 samples.
But it does not mean timing resolution doesn't also depend on sample rate.
Rather, both affect resolution, and not just timing resolution.
This is evident from very high sample rate 1 bit streams like DSD, which have very low sample width, yet produce excellent resolution both in 'levels' and 'timing'.
As a general rule, you can reduce sample width as long as you increase your sample rate enough, and the other way around, you can reduce your sample rate, as long as you increase sample width, but within reason. Your sample rate must at the very least be high enough to cover the frequency range you want to capture (ie, 2x the highest frequency you want to deal with).
@@c128stuff I agree with your comments. What I was referring to, considering the first audio CD players, was the analog filters that were used at that time to cut down any frequencies above 20kHz, in order to eliminate aliasing: i.e. two frequencies passing from the exact same samples you have recorded, when you attempt to to play them back. From a pure mathematical point of view, having a window of sound sampled at 44100Hz, can contain a spectrum of frequencies up to 22050Hz if you perform a spectrum analysis, but the same spectrum mirrored after the 22050Hz is also satisfying the same samples. Even though our ears obviously can’t hear those high frequencies, the differences between them create harmonics in the audible spectrum, and that’s why you want to cut them off. So if you wanted to have a frequency response from 20Hz to 20kHz from that sample, you’d need an analog filter which is flat in that frequency range, and then drops dramatically from 20,000Hz to 22,050Hz. That was the one introducing that phase shift I was referring to back then, considering the analog electronics of the ‘80s too.
So two ends of the range are Niquist sampling with infinite bits per sample giving infinitely accurate quantization , and on the other end is infinite sampling rate with 1 bit per sample. And we choose to work somewhere in between, giving best quality / size tradeoff
The thing that is nice with higher sampling rates is that you can line in feed directly a very good 44,1 khz recording and upsample a higher signal thru a ok Dac.
Then pass the output to a console and finally to the recording input of your sound card.
The console is really interesting because here you can modify the sound and induce a surround effect to the dac output signal and makes it actually produce audio to those frequencies above 22khz all the way to 48khz, within a 96 khz recording sample.
I have digital samples of each pipes sounds of my vitual pipe organ recorded at 96 Khz/24bits. Up sampling from 44.1khz won't give the same detailed timbres as shown by waveforms.
When recording with High quality sound card, the ADC is in hardware so it won't affect much CPU cycles.
OP says he can't hear the difference. But OLDER people like him has "presbycusis". That's why opera is mostly attended by older people since young people finds opera irritating.
Thanks for sharing. Can't say I understand all of them but have some ideas of it. Someone told me before that 44.1 is the best.
At 192Khz reel2reel archiving, it also captures the 50-70Khz BIAS oscillator faintly from the original recorder.
Now this oscillator drifts as the recorder warmed up (tube era), either its mass, or surrounding environment.
If it was stable, it could have been a better source to calibrate the motor/mechanical speed shifts shift on original deteriorated tape, but many times there is faint 50hz hum, assuming that was more accurate, I use that.
Thus, the Plangent Process works so well at correcting wow and flutter.
Thank you sir!!! A good video to show the kids ( I've been making them read Mojo Audio's: 24 bit Delusion).
Virtually all ADCs of the 90s were 1-bit 64fs. In other words, DSD is just recording the first stage of converting to PCM and does away with the need for brick wall digital anti-aliasing filters, as well as reconstruction filters on play back. Nowadays, most quality converters are 2-bit or even 3-bit, at 128-256fs but the real world performance isn't that much greater than can be achieved with 1-bit 64fs ADCs. The main difference is you can get away with not having an analogue anti-aliasing filter (usually 5-pole) in front of the ADC.
Wauw, there is some deafening truth on this channel. Love it!
As I remember being told back then, that the record companies where affraid for their buisness when the DAT came out, since you could make a digital clone of a CD, hence the 48kHz. For a long time it was impossible to make a digital copy to DAT from a CD-audio, or record analog to a DAT on 44.1 kHz.
Most of the video-broaadcast is still working on 48 kHz, and they won’t upgrade to 96kHz anny time soon, since the benefit of audioquality is minime compared to cost of doubling the data rate.
But for acoustic music production I would strongly advise to record in 96 kHz 24 bits, mix in the DAW in 32 bit flow or better, and only SampleRateConvert after mixing. SRC can induce very nasty audible side-effects.
Imho the benefits for recording and mixing at 48 kHz are lost with SRC if you need to go to 44.1 kHz for CD-audio.
Our ears can only hear maybe up to 15 kHz or so, but what about the harmonics created by insruments above 20 kHz? Take two oscilators, one at 25 kHz and the other at 30 kHz. Now vary one of the two, and listnen to the harmonics you can perceive with your ears. I presume most fo these harmonics are lost when recording at 44.1 or 48 kHz.
24bit 48Khz for recording, mixing, mastering
(24bit 96khz only for raw recordings and source archive purposes;
for special applications, scientific, we might go for 32bit 96kHz but those are not really for music)
If we have 24b\48kHz, then DSP (digital sound processing) should include 'over-sampling' (×2 or ×4 the orig. freq. 48kHz in our case).
Record near 'hot' levels (test with low + percussive sound for worst case scenario of a small 'headroom VU' - eaten 'volume units' by the low freq. test sound, and louder Peaks - the percussive hit), if clipping occurs for a few samples DO NOT overthink it - they can be restored in post-recording\pre-mixing production!
Appreciate the content and humour
Great video and I agree with mostly everything you said. Whenever I multi-track, I use 24/48 as it's senseless to use anything else for the reasons you mentioned. When recording my vinyl records however, I always use 24/96 as that is usually what the master used in the process was given to the cutting engineer, on recent releases anyway. Anything vintage on vinyl I just use 24/48, but new releases on vinyl are truly magical and I don't want to miss a single harmonic.
"I always use 24/96 as that is usually what the master used in the process was given to the cutting engineer"
No. normal Vinyls were not even close to that. Commonly used cutoffs were in the range of 17-19kHz to not stress the cutter unnecessarily, with dynamic range rarely touching 16bit.
Recording with higher fidelity for preservation is fine, but don't kidd your self into thinking that vinyls are even close to being capable of reproducing those numbers.
I think they meant the masters that were given to the mastering engineer for vinyl were 24/96, not the final output. But to take away from your point, but interesting to see that the final output is so more constrained due to vinyl, crazy! But vinyl can still sound amazing.
My late father (who did a lot of voice-over work) was acquainted with a prominent sound engineer (in Australia) who did a lot of work recording/engineering orchestral music (classical and sound track) who 'bounced' multi-tracks to stereo _through an analogue desk_ (back to digital) because he found the results to sound _considerably_ better than using a DAW's 'sum and differencing'. A digital zealot and reductionist like our host will, of course, find the idea preposterous. BTW, 2822.2 kHz is the sampling rate for 'DSD64' as it first appeared nearly 30 years ago (as an industry archival format, not an "audiophile" one - most people seem to be ignorant of this) higher frequency/resolution DSD128, DSD256 (and even higher still) have long since been available (thanks to those nutty audiophiles, presumably).
As a producer I work with a lot of legacy 44.1kHz 16 bit source material (CDs).
My DAW is set to work with a resolution of 24bits, but I tend to stick with a project sampling rate of 44.1 kHz, beholden to the idea that converting from 44.1 to 48 might create artifacts in the audio, and not changing the sampling rate is best practice.
I suspect I might be dwelling in the stone age here, but I'd like to try and catch up!
We should consider this debate over, and start discussing lossy/lossless compression formats more! Audiophiles are saying it's all "mp3" but modern compression can achieve way better sound while using less bandwidth/storage. The Opus compression kinda blew my mind, as it is maybe 3x more efficient than mp3 (and used by youtube for its audio).
Should we? Says who?! 😂
@@fernandofonseca3354 I say. Opus is indeed amazing.
@@fernandofonseca3354 says me, an audio professional and huge nerd
Good, then go ahead and by all means hijack the thread! Enjoy your christmas day. Next!😁
Great video, it's drove me 30 years back, when I was a student in university. I clearly remember "twice-samplerate-difference", Kotelnikov theorem. All the rest is much simpler: more bits (16/24) means more accuracy for DAC, 48kHz is more often than 44.1 :) Just subscribed, and thanks for your job!
More bits DOESN'T mean more accuracy as long as proper dithering is used. It means lower noise floor. Consumer audio needs about 13 bits of dynamic range for the noise floor to be inaudible in all reasonable listening scenarios. Clearly 16 bit is enough in consumer audio while 24 bit is ridiculous overkill.
did you hear 1bit or 8bit audio? :) No noise at all :) @@pojuantsalo3475
I don't care about bits or KHz or sampling rates. All I know is that digital music sounds so much better than records and tape. While I still love my retro component audio system from the 80s, my "go to" listening experience is my tiny Sony MP3 player and a pair of Logitech THX speakers and subwoofer or a good pair of Sennheiser headphones. No hiss, no rumble, no clicks or pops, perfect channel separation, bass I can feel even at low volumes, highs so crisp I swear they could break glass.
Mp3 is mainly sacrifying high frequencies (to obtain its low bitrate of max 320kbps). With 320kbps you can loose 60% of the information in the audiofile..... that s a lot and you would hear that on any decent audioset (not a tiny mpr3player).
@@thepuma2012 i think he wants to say that even an mp3 sounds better than any old audio medium from the 80`s
which i agree with tbh. I mean there are some good Vinyl Decks but they dont sound „better“ they just sound „different“ imho and
i have a decent Deck and enjoy Vinyl once in a while but if you compare for example bass heavy, (relatively) modern music
on Vinyl and the same Song-Track on mp3 320 its day and night. You cant push Vinyl too much when it comes to stereo information. You need to stay much narrow-er in the lows and in the highs on productions meant to be cut on vinyl.
Vinyl simply cant keep up on a technical level.. But i think it`s predominantly a matter of taste.
Tape also sounds cool and i basically grew up with vinyl and tape but technology has simply moved on. I had the best and most expensive sony walkman (the ones covered in chrome) in the 90`s and then Minidisc came and i had a friend saying „The worst Minidisc walkman is tenfold better than any tape walkman“ and at first i did not want to believe him until i compared them. It was simply day and night
To your "60% loss of information“ i can say that i have encountered very very few people who can actually hear a difference
of 320kbit to Cd resolution given that you will use a good D/A converter for both.
My cousin for example will throw all my arguments and say vinyl sounds better. Its a funny subject thats for sure
@@saardean4481 ok. I understand all you say about the vinyl and 80s audio. But not hearing mp3 to CD difference with a good DAC i find surprising. Indeed a funny subject as you say 🙂
@@thepuma2012 Surprises are the salt of life. I am worried more about the increasing lack of dynamic range and abundance of distortion in modern music rather than mp3 artifacts. As for your reference " But not hearing mp3 to CD difference“ maybe it would help to read again. I said " i have encountered very very few people who can actually hear a difference
of 320kbit to Cd resolution“. If you on the other hand have encountered many people that can easily distinguish quality differences between 320 and cd in a blind test then you are a lucky person surrounded by gifted people. You should treasure this 😉
@@saardean4481 Absolutely agree about that "loudness war" issue! I do have a Blu-ray player, but i refuse to buy any audio blu-ray because all of them have compressed DR - even so that an LP version of the same music can have more dynamic range than the disc version. Also on stream-media that problem.
I believe that some people used the 48khz rate because that is what film and television production requires. It fit into their standard of frame rates. SMPTE frame rates, etc.
For me, it is probably the quality of the sound engineering.
If it is recorded in 24 bit 44 or above, it means they more than likely care and it has gone through a more modern process and mastering.
The difference in mastering quality is massive between, say, an 80s song and a song in 2020s.
A lot of the cd era 90s pop was mastered so poorly.
The cds sound metallic and harsh, unlike the older tape process.
The tape process was warm but compressed.
As an audiophile, it is literary hard to listen to older cd mastered music anymore, most likely due to the bad mastering processes used.
If you find them in 24 stream or flac and it is remastered, it sounds very different, not because of the bits but the process of mastering with modern day tools.
Would love to see a video on this... ❤️
I chose 48kHz, for technical reasons related to streaming. I can run run all devices at 48kHz, but not all of them at 44.1kHz. And dealing with multiple sampling rates can be a pain in delays caused by conversion. So while 44.1KHz is enough on the audio side of things, 48KHz is good on both, audio and technical handling.
As for the "utility" of it, well to each his/her own. I certainly see a use for a medium capable of storing more information than apparently necessary which is down to documentation/curating purposes.
If you record and mix in 44.1 or 48, then I suppose that's typically OK provided that the digital effects are oversampling to remove any aliasing from the application of those effects?
Correct. Filtered on the way in, filtered on the way out.
What makes the most audible difference is the objective distortion at low-but-perceptual levels (down to, around, -60dBFS). This is why multi-path architecture excels over any legacy single-path design.
Gotta love audiophile Phil.
I feel as if I've known him all my life!
I know he is just - as it were - your alter ego.
So every time I see him, In my head he's audiophile Shill.
I don't think the question is what sampling rate is overtly better but how the audio is interpreted and reproduced. With a lower sampling rate information about the shape of higher frequency wave forms is lost. For instance consider a square wave at 16khz. At the simplest a square wave requires 4 points. If you sample at 32khz that leaves only 2 points to depict a square. This is a problem even at higher sampling rates (up to 64khz for our 16khz square example), due to the waveform being represented by too few points. This translates into different frequencies in the Fourier transform that make up the signal, which change the harmonics and how the tone is perceived. It's possible to argue the limitations of a speaker system and general consumer use. However, to ensure the original intentions of the sound are most accurately portraied the highest fidelity source should be used.
@nicksterj On most systems sure, speakers will dither the waveform. But the idea isn't what your speakers can produce, but the fidelity of the signal you provide them. Work backward from 44khz, the best possible square signal is only 11khz. And for a 32khz sample rate, 8khz. Now also consider how the fidelity of any audio signal can be expressed as the difference between the sum of infinite sinusodal frequencies and what the source is capable of reproducing. In any case lower sampling rates always reduce the ability to reproduce a signal.
ua-cam.com/video/spUNpyF58BY/v-deo.htmlsi=TREI8CXTpvmN0lLx
The critical part in the whole chain from analog sound to samples on a cd is the anti alias filter. What is the cut off frequency, how steep is it and what is its phase spectrum. This is an analog filter, applied before sampling. So how does this work in practice, that is what I like to know.
Considering so many speakers use metal dome tweeters which break up with a huge peaks (probably 10 dB or more) above 20 kHz I happen to be believe 44.1 kHz most be best. Not that I believe I or most others can hear or sense it. But why give the dog or local bat a migraine attack? We must be nice to the animals 😊 While writing this it came to my attention that while bats can hear frequencies up to 110.000 Hz the Beluga Whale can hear up 123.000 Hz. It must be quite a challenge for our whale audiophile friends to find underwater speakers to serve their needs 🤔
Greetings from Denmark
Since modern AD and DA converters work with pulse density (1 bit), the precision must decrease with increasing sampling frequency while the chip master clock remains the same. However, at 96 KHz you have twice as many samples with less precision and an anti-aliasing filter with less phase rotation and better transient response. I would consider 96 KHz to be the best compromise for real analog sources, although for my ears it ends at 18600 Hz. I've never been to a disco.😉
Audio Phil is my new hero...
Send pics when you have your double denim tux.
24 vs. 16 bps guarantees 2-3 times as big FLAC files. What I often do is bringing the bits per sample down to 20-22 (so there's zeroing the LSB, technically still 24 bps), via dithering. So a lot of my audio ends up as 48/20.
Storage is cheap, sure, but with high sample rates and high bps space goes out of hand.
Converting FLAC source to target devices, I always go for 48/24 (even upsample 44.1/16 with SoX), because most devices handle it natively - no resampling on the fly.
i don't wanna get into all this samplerate talk rn but 32bit is surely useful in production, cause the resulting wav files don't clip at 0db, so they are easier to work with than 16 or 24bit wavs. if 16bit wavs had that feature I'd likely produce at 16bit because that's what my soundcard supports
32-bit float would have been a bit off topic for this video, but for files created during production, perhaps to share with a collaborator, then yes it's the way to go.
I will never get tired of listening to Audio Phil.
I had trouble putting my finger on what it is about Phil that I like so much but I think it's the confidence he exudes while he speaks his over the top audio-science jargon. He's like a skillful politician who makes you want to believe that what he's saying is true.
two-channel 16-bit PCM encoding at a 44.1 kHz sampling rate per channel, with four bit ovesrsampling and error correction
from heaven is fantastic.
Incredible I first thought it was Paul McCartney speaking here : and you even sound like him ! 🎉
I deny everything ua-cam.com/video/aB3JNivlMnI/v-deo.html
Greetings David! What you propose is always very interesting and coherent 👍👍👍👏👏👏
I am an audiophile and an old professional speaker who has been through many recording studios and I have directed radio and audiovisual post production, starting me in the era of open tapes for professional recording.
I don't have your professional knowledge but I have always been interested in the subject. I have read some articles and watched some videos, which comment on the harmonics of each fundamental. They mention that within the audible frequencies of 20Hz to 20 KHz, there are harmonics at different frequencies, which can even reach from 49 to 225 KHz. In one of those videos they showed comparative measurements between CD reproductions and streaming, noting in the latter provided by the web, the lack of harmonics in most of the known platforms and only a couple of them reached almost 50KHz in their harmonics. while the CD did it up to 225 KHz.
Of course, on the other hand, there is also the issue of the quality of the original recording, as you always refer to. But obviously, even if young people do not hear frequencies higher than 20KHz, an amplifier and speakers that have a frequency response much higher than the audible one would, in theory, be very beneficial for those harmonics that end up configuring each fundamental wave. What's your opinion about it? Greetings and thanks for the topics you present.
I think it's useful to put a figure on frequency response and if there are humans, no matter how young, who can hear 20 kHz, clearly 20 kHz is only just enough. I'd probably double that to 40 kHz, as in the spec for hi-res analogue. Any more than that is a) definitely wasted, and b) asking for trouble.
@@AudioMasterclass Thank you very much for sending me your opinion, you are very kind 👍👍👍
Is it not: For finished audio files, just to the listeners the resolution makes more difference than the sampling rate when the master properly is converted to 44.1.
But for music production and recording the 96 kHz is required to minimize the LPF filter distorsion and rounding and all of that since the audio will be processed and maybe converted more than one time
48K for tracking - as my channel expanders are ADAT, I'd lose half the outputs on each 8 channels at a higher rate, but 48 still sounds great for me. One benefit - as I understand it - with higher sample rate capture above 48, is the files work better with elastic audio processing ITB.
The last sentences definitely for me
Probably the later. But to the serious stuff. I don't really know if I am correct, but Hemholtz theorized a bandwith--acoustic and non-recorded of course, of around 46khz, because from his experiments he felt that young teenagers could hear up to 23khz, or it might have been 22.5khz. So he wanted a frequency rage twice that-- 46khz-- so that all harmonics below 22.5khz would be able to interact with the information above it and produce combination harmonics in the audible band. Say 20khz combined with 25khz would produce a combination harmonic of 5 kHz, though of course at alower level than the original two tones.
So you need to get to 46khz, so that 46 and 22.5--the highest audible frequency--would produce a combination or difference tone of 23.5khz, which would be out of the audible range of hearing for a young teenager with no hearing damage. That was his theory and was why he wanted a bandwidth of 46khz or so. It is also why some manufacturers in the past, especially Harman/Kardon, and I'm sure some British brands, emphasized large bandwidth amplification and the good reproduction of square waves, which is an indication of bandwidth. Essentially, they wanted flat response up to 46khz or beyond. The great Stuart Hegeman engineered some design that went up to 300khz. And the square squarewaves measure by hi-fi magazines showed that. This was in the 60s and 70s. Harman/Kardon still designs this way even though the Harman group, which included a large professional and recording division, and notably, AKG of Austria, was purchased by Samsung, the giant South Korean manufacturer, about 8 years ago. Sad but true.
But getting back to the central point. I know I can't hear that high, and I'm not really sure if this matters or if Hemholtz was correct. But the issue is interesting, or intriguing, as the British would say. Maybe we do need a bandwidth and a sampling rate of 88.2 to get all those harmonics into the audible range. I really don't know.
As for bit depth, "whew, that's a complicated matter". I guess it can get you lower noise and a larger dynamic range. I can't say if it is of practical use.
You need high sampling rates to reproduce transients as these contain multiple harmonics of the original frequency. The harmonics in isolation are above our range of hearing but the effect they have on the shape of the waveform is detectable as a very sudden transient. As for bit depth it is just not about dynamic range but resolving power and again to reproduce music which is not a smooth sine wave but a messy waveform loaded with harmonic variations then 24bit does a better job.
Don’t forget the lofi of us out there .. I’m sampling at 13bit (Ensoniq eps-m sampler) 28-39khz and it sounds fantastic
Yes this is a thing for music creation. As well as low bit-rate, I rather liked the aliasing in the Sequential Circuits Prophet 2000 and was rather disappointed my Akai S1100 didn't have it.
Interesting video. Thanks for that. Making a jump to the real world. Our hearing is analog and we use digital as a bridge. I went to a Roger waters where he performed " the wall" and during the act they build up a wall. at one moment you don't see the band anymore and you are looking at that wall. I made the joke that the band is likely backstage having a drink while we now listen to a tape. But that's in essence the best thing that can happen. If I can't hear the difference between a live recording and the playback I'm 100% satisfied for I'm there when I close my eyes.
And what comes closest to that ? is it still tape ? is DSD the best thing ? or PCM ? I don't know for I don't have that reference.
Thanks for this - can you do a segment on this for bluetooth codecs? They're all over the place with aptX lossless, LDAC (which seems to not actually ever achieve its highest bitrate due to RF interference), LHDC (which no device supports), AAC (black box), etc - there's no clear & intelligent source of truth on this subject
Bluetooth is a topic I've been putting off for a long time. I might put it off a while longer. I promise I'll cover it when I reach 100k subscribers. (That will give me some breathing space)
I used to mix in 48Khz, while this maybe enough to convey the mix to any listener, the math behind it or rather said the simulation of analog gear will be less acurate compared to 96Khz
And for those who work like people in the analog domain, comiting to a sound or dymamic feel is mandatory to keep the computer fast. I always commit virtual instruments with loads of plugins and bounce them to an audio track. If I could i would rather mix on 192Khz that is what most commerial studios do and they are not wrong. Mr. Neve has said this and he is right, there is an psychoacoustic ellement in the high frequencies that is experienced by listeners, this was one of the revelations mr. Neve had during his life.
This is new to me that the higher rates are purely for capturing higher frequencies. What about the higher capturing fidelity of the waveforms? That's always a good thing if the pure signal dictates the sound more, and not the DAC. It isn't very critical of a topic looking at pure single frequencies, high or low, but in the case of a complex sound, the higher sampling rate, the less it all becomes white noise in the end.
Recording is done at the Nyquist rate so double the sample rate. e.g. 44.1 recordings are sampling at 88.2. Doing so already preserves 'all' the signal you input upto the sample rate. Nobody can hear above a 44.1 sample rate so recording information above it with even higher sample rates is pointless, unless your analogue converters do actually sound better at higher sample perhaps and/or the conversion to lower sample rates sound better than just recording at them. e.g. a drum being pitched up or down will pitch down more cleanly if you recorded at higher sample rate etc.
In my experience 96k gives less sibilance, less need for de essing and corrective high end EQ, better sounding pitch shift and warping. The best place to listen is in a vocal or a snare. If I can’t hear it immediately, after processing and limiting and on multiple tracks it becomes night and day. If I need to save space with other tracks il make sure the vocal and snare at least are 96
The reason I heard for the 44.1Khz was because the human hearing has a frequency range of 20Hz to 20KHz and, according to Nyquist, to convert an analog signal to digital without any loss you have to sample it at twice the rate of the maximum frequency.
By sampling it at higher than 44.1KHz you're just losing storage space.
Yeah, its due to that as well but the exact sampling frequency apparently originates to that audio recorded to a video recorder that he spoke about, i havent heard about that before this but it was common earlier on to use the same formats for better compatibility betweens systems, nowdays things are often far more flexible. I agree, 44.1KHz is already "full res" anything over that is mostly wasted space and CPU power, i record my band in 44.1KHz 24 bit in my DAW for our albums.
Oh, my hard drive stuffed full of 96 vinyl rips begs to differ it's just a waste of space LOL.
Well, there are benefits of working with higher frequencies and higher bitrates for a DAW internal audio engine, allowing more accurate multiple tracks mixing and audio processing.
Obviously afterwards, for whatever actual output format, 44.1/48 kHz are good enough.
Working with higher bitrates also allows a lower input latency while still having a decent/large buffer size (avoiding buffer under runs).
With the newer CPUs, I feel 96 kHz or even 192 kHz shouldn't be an issue but unfortunately there are plug-ins (VST64/VST3) which won't work if your DAW is set to such high sampling frequencies...
I found myself having to limit my expectations and stick to 48 kHz with a pretty high buffer size (2048 bytes) if I want to be able to stack many synthesizers with no crackling...
The input latency is pretty bad but I can live with it.
I read not too long ago in some forum that 96 kHz sampling is the thing if you are thinking on releasing vinyl records (analog). I don't know if there's any truth in that statement or it's just BS as per usual.
Discussing 32 bit float soon? Just subscribed to your channel. Thanks for the video. Cheers.
32-bit float... Yes, yes, yes, then no. More in a future video.
I recently purchased the S.M.S.L. su-9 pro, because I wanted to be able to decode DSD to DSD512, and 24bit/192KHz for playback with my nearfield listening system. Trying not to be the audiophile that Alan Parsons refers to as listening to the equipment, and not the music, the main difference I think I hear, is not audio quality but time correction, and that probably has nothing to do with the sampling rate at all. My playback is though a 1975 Pioneer SX434 into Wharfdale Diamond 9.1's which probably produces the quality of sound somewhat equivalent to my aged hearing
semoga sehat selalu abah 😊
High sampling rates for playback are insane 48K is the accepted norm for all platforms and is a happy medium but even 44.1 is fine for playback in the final product. But for music production using high sampling rates like 96K is a bulwark against aliasing distortions. Its not about the bandwidth and how much of that which you can hear. But high sampling rates such as 96K pushing aliasing distortion out of the hearing range that would otherwise creep down to the audible range at 441 or 48K. It helps when using digital processing like saturation, and anything that adds harmonic content to the signal, it also gives some EQs the ability to not cramp. I heard a professor from ETSU's music school describe sampling rates in a very interesting way once. I wish I could animate it because its a great visual. He said imagine standing in front of a 10,000 feet tall concrete wall with a concrete floor and you are standing there with a tennis ball (aliasing distortion) in your hand. Imagine standing 21.5 feet away (44.1K) from the wall and throw the tennis ball as hard as you can and where ever it lands that is where aliasing would be heard in the hearing range. Now move back to 48 feet away (96K) and perform the same throw. Most likely at 21.5 feet the ball went over your head and landed behind you somewhere, but at 48 feet the ball is less likely to even make it back to the spot where you are standing. I thought this was a really great visual for describing aliasing.
@@RockManEnough Oversampling in converters can help give the filter a more rounded shape, but the fact remains there is more to a session than just recording the audio through the converter. the antialiasing filter is still in the same place. So no matter how many times the chip will oversample x128 or x512 the filter at 48K will still be at 24 kilohertz on the output leaving the antialiasing filter in the same place same place but with a more favorable shape. Oversampling isn't a replacement for using better LP filter placement. The type of filter used as well is highly important. Lavry explained this in the sampling rate white paper that was published nearly 20 years ago when they were building their next generation conversion chips and they explain why converters with oversampling can mask many of the failures of digital conversion filters. Its one of those deep audio subjects you can't really have a good conversation about in the UA-cam comments section. But saying there shouldn't be aliasing at any sampling rate for converters is true but we don't just record audio and thats it. We also use digital tools and effects that will work better at higher sampling rates. Time stretching, Tuning, Distortion, Saturation and Pitch algorithms all benefit from the highest sampling rate possible because this gives them more data points to work with. But in the end if it sounds good it is good so thats the great part about choice if a person wants to work a lower sampling rate they can make music at 44.1K and if other people want to work at 192K they can. I work at 96K because its the best spot for me but thats the great thing we can work how ever we choose. Cheers, Have a great day!
The guy at PS Audio said on UA-cam that they have started using DXD when making recordings which they then transfer to DSD. That is, significantly higher frequency than 48 kHz, they have a record label called Octave Records
Note that guy also answers questions from people on youtube . If I understood correctly, he is the founder and CEO of the company
If you are producing music, the delivery format defines the sampling frequency and bit depth.
24/48k is the norm these days.
Multitracking with 16 bits will save on disk space at the expense of a higher noise floor.
If you’re applying any non-linear processing then higher samples rate (or oversampling) will help combat any aliasing.
It is absolutely true that 16-bit has a higher noise floor. However it's likely that most or all of the faders will be lower than 0 dB so there's less noise than there would otherwise be. If you're mixing to 24-bit of course.
I very much enjoy your educational videos. Would you consider making a video about harmonics or ultrasonics? Frequencies from instruments higher than humans can hear, that may or may not add to the perceived sound of music when played live or when listening to 96khz or above recordings through resolving audio hardware and or Hi Res headphones. I keep reading and hearing of this perceived sound that subconsciously makes the sound we hear more real. High frequencies that our body can sense but not necessarily hear, like sub bass below 20hz. I'm wondering about the benefit of adding a pair of super tweeters to my home hi-fi system. Would love to learn your thoughts. Zach
I have recorded with 24 bits at 96Khz for years, not because I can hear a difference in sound, but because I believe that I get greater headroom and much less aliasing with this choice. And as I typically work with about 100 tracks in a song, reducing the problems created by lack of headroom and distortion from aliasing seemed like a good idea, even if there seem small. It costs more to be able to do this, but I do it anyway. Maybe I am wasting money and effort ?? Thanks!
I love 1Khz as a tone. Smashing.
Thank you for this topic.
1) Sampling rate for Capture:
I do many conversions and audio restoration from analogue LP and RTR masters to digital for the purpose of remastering.
Spectrum analyzing the musical signals of an typical recording reaches high frequency of around 40Khz and better quality reaches even up to 60Khz (!)
If I want to fully capture the whole information I need to use a sampling rate of 192KHz.
However, the 192KHz is a difficult task to handle in many aspects' including the A/D performance (Signal/noise) and from practical reasons I use only 96KHz/24 Bit Sampling rate for the capture and for the post processing.
2) 20KHz and above musical information:
The signal level may be low at these frequency but the energy (Power x Time) is high. if we would like to fully preserve the full musical information we must sample the signal at high sampling rage.
3) Can we hear the Signal above 20KHz?
We can't Hear or identify tones (Frequencies) above 20Khz, but we are capable of feeling the presence of high energy of high frequency when it's envelope is modulated by an audible signals (similar to the principle of how we can recognize a dithered low level audio signal).
4) To my experience the 96Khz sampling rate is a must from capture till reproduction.
5) D/A Filter Slope:
When sampling a 20KHz audio signal at 100KHz there is an aliasing effect around the 100KHz from 80 to 120Khz where the 80Khz artiact is generated by the 20KHz signal and 99900Hz is generated by the 100Hz signal.
When Sampling rate is 48KHz the Alias of 20KHz is generated at 28Khz. There for the LPF of 48KHZ must Pass 20KHz and Must Stop at 28KHz. it is a very challenging slope for the LPF, which presents heavy phase distortion within the audible range.
6) 1 bit: provides a very accurate linearity for low level signals capture and reproduction. has lower bit resolution at higher frequencies as well as higher noise level than a PCM.
Thanks Again.
Hold on, you're saying we can hear unhearable ultrasonics when they're modulated by sonics???? What are you waffling about ?
@@NamelessSmile a high frequency can swamp a gain stage -- you can't hear the frequency that's causing the distortion, but you can hear what it does in the audible range.
of course you can get high frequencies off of tape: the bias signals for most decks are up in the 30kHz+ range, and if you can recover them, you can correct for wow/flutter. (to good effect, as well!)
@@poofygoof yeah that's true, doesn't mean it impacts the passband of a filter it can't get through
Recording to 60kHz? Bunch of baloney, also I don't remember Fletcher-Munson curves going nowhere near past 20kHz.
funny thing is some 1980s famous tracks were recorded at 32khz 16bit pcm lol
Did he even answer what the title says - 24 bit or 96k hz which is more important? He just went on and on about sampling rate.
Anything over 40khz enables one to perfectly reconstruct signals below 20kHz. The problem is impact type sounds ( like percussion) , for one, exceed that 20khz bandwidth and even though you cant hear a 20khz sine wave, you can hear the sharp rising waveform of a drum thump as well as “distortion” on real instruments. Then also, the way the signal is reconstructed- rarely do the compute inverse laplace transforms etc - they just interpolate. Hence the divergence of opinions
Distortion and wild frequencies make their way up past 20khz into ultrasonic ranges. You can get mics like the Sennheiser MKH8000 series that can pick up 50khz. Anything above 20khz isn’t just noise. There are harmonics up there just like there are resonances all over. Sure, it’s not too practical. Not a lot of people recording bat sonar pulses and slowing them down, but they do exist.
88.2 is probably ideal. You want to be able to capture at least upper harmonica and resonances that make distortion sound brillaint. 192khz and 176.4 are great for sound design. You have to think about if you might need to pitch shift something.
If you’re just going to be recording nursery rhymes and polka, you’re absolutely safe at 44.1 and 48
DSD goes at some crazy high rates 5.6mhz and 11.2. Might even go up to 22.4. I can’t remember. And yes it sounds amazing.
@nicksterj I have qualifiers in there like pitchshifting, and "Sure, it’s not too practical. Not a lot of people recording bat sonar pulses and slowing them down, but they do exist."
What you're saying is practical, but there are esoteric reasons beyond, which makes it worth it. Especially archival purposes. It's like saying "my jpg or bmp files are perfectly fine because they have all of the colors I like in it already. I'll be deleting my raw files" No, the raw files have extended data that you can access and remaster later.
Try recording a lot of distortion. I can hear a distinct difference on my system. At 44.1 and 48khz it sounds rolled off. Not just guitar distortion but going through something like a Thermionic Culture Vulture or something that can introduce a lot of pentode/triode distortion into a layer. It gets tricky at high frequencies you're dealing with a lot of micro distortions essentially giving you high freq square waves. They'll capture at 44.1/48, but the math gets a little fuzzier.
Also like the guy said in this video, at higher frequencies you get a better rolloff.
I think going too high in frequency like 192khz, unless you're doing sound design, is a bit overkill.
The sweet spot in my experience (probably going on 27, 28 years) 88.2/96 is the sweet spot. I can see the arguments for going 44.1 to save on processing power, but computers can handle it these days.
that's not what's being addressed here. It's if it matters. According to that, it does@nicksterj
New advances in computing are looking at replacing digital processing with analogue processing -or digital analogue hybrids- due to being more energy efficient, etc. Well sound processing and storage may go back to analogue again?