The biggest reason why you would record at 96,000 or 192,000 would be if you intend manipulate the recording. If you are recording singing or guitar 96,000 will allow you to pitch correct your singer and or your guitar solo to some degree in which it will then cut your frequency in half. You cannot auto tune or pitch correct a 44,000 sample and keep it at 44,000 thus you would no longer have cd quality after the correction. with 192,000 you range of manipulation flexibility and slow down possibility should be even greater. I would encourage anyone to record at the highest frequency possible because at 44,000 you will have no wiggle room to fix or manipulate slowdown harmonize etc etc your work. You should always keep your master recording un manipulated and make a double of it to manipulate it because you may want to revisit your stems it in the future possibly for other work.
The main problem here is that if you don't downsampling to 48/44.1khz your audio will be automatically downsampled and ruined by streaming service that will not let your audio play at the original sample rate, however the max for playback streaming on youtube is 48khz so if your song (original) was higher, we will never know :(
UA-cam resamples everything to 44.1khz. Read their specifications. The tragedy here is that most of the people who use youtube or a music streaming service (44.1khz) will use their smart phone or tablet to do so which have a native internal sample rate of 48khz as it's a media device. UA-cam is a video platform and should have 48khz as their sample rate as that is the video standard sample rate. But this world is ass backwards and youtube sticks to 44.1khz for whatever idiotic reason. So when the blue ray disc of a movie comes out it will be it's original video sample rate of 48khz, but once uploaded to UA-cam it will become 44.1khz and the device you watch it on will have a native sample rate of 48khz.
I tend to use the Saracon though I agree it does have a sound to it. RX tends to just not have a sound at all whereas the saracon has an emphasis of sorts around the nyquist
Wondering if I'm following correctly: are you doing this process before or after you've mastered? And wondering how you go about capturing the file to the desired sample rate (the process you mention at 2:00)?
Really good video on this topic! My preference is to pick a relatively high sample rate and bit depth and stick to it as much as possible. If I have to convert the sample rate, I try to do it by a nice clean integer factor. This is usually going to create two common families of sample rates, CD-based, which is 44.1, 88.2, 176.4, 352.8, 705.6, 1411.2, etc… and DVD-based, which is 48, 96, 192, 384, 768, 1536, etc… My preference is to pick a relatively high sample rate from one of these two families and stick to it. If I have to convert it, I try to stay within the same family so that it can go up or down by doubling or halving. When converting, you use a brick wall filter to cut off frequencies above the Nyquist limit of the final sample rate, which is one half of the final sample rate. You then cut out samples at regular intervals using decimation. For example to go from 96 to 48 you only use every other sample since you are going down by a factor of two. 192 to 48 uses every 1st out of 4 samples and skips three for each sample cycle. If I have to convert by a non-integer factor, 48 to 44.1 for example, I just try to use the best interpolation algorithm that I can find to get the audio as close to the original analog continuous waveform signal as possible. Obviously if you need to convert something like bit-depth/sample-size/word-length you round it to the nearest integer. If you need to convert the modulation and/or demodulation method you use an algorithm to get as close as you can. Obviously like you mentioned different algorithms, software, firmware, hardware, tools, AI models, artificial neural-networks and/or equipment can yield better or worse quality results, so it’s definitely a matter of finding a good one like Isotope for example.
On some of my projects i use samples, which most of 'em are 44k 24bit....but i like to record my hardware synths at 96k 32 floating point. If the arrangement has most of the tracks recorded from hardware synths, is it wrong to upsample those loop samples coming from libraries at 96? Or record the hardware synths at 44k as the lowest sample rate in the mix?!?
That's what I also want to know. ROMplers like Omnisphere and Nexus are 44.1 as well... And I use them a lot... Should I bounce them as audio track and upsample...
I know things have improved since 44.1/16 times but I found back then that when a waveform is processed too many times it develops a peculiar "sheen" and loses presence. The "generations" problem of analogue tape seemed to have gone away but that didn't mean we could apply more EQ to fix the EQ that we'd applied to fix the EQ. Obviously at todays resolutions that's less obvious but it still makes me think it's wise to avoid unnecessary math.
Thank you so much for the video! May I ask? Is thiis out of date now in terms of SRC as I see logic, ableton and cubase have upgraded their SRC and charts show there to be no issues now?
This is so intriguing. I'll definitely be testing out streaky for a master soon. I love the detail he's taking into trying to get the absolute best sound possible.
I'm listening on headphones (Focal Listen Pro's) and I can actually hear it! The original sounds better...the depth and clarity. I didn't think I would be able to hear it on UA-cam but I can...the silk is in the original but for me I can't hear that in both of the conversions. It's super subtle but there's definitely a difference if focusing on the detail. Like the F1 drivers making changes to win the race with milliseconds...mastering is the equivalent...so in this context I can hear it and it does make a difference. It's a shame that it needs to be down sampled...it definitely sounds the best in it's original form.
I isolated drums and bass of a track. The track is 44100 k 16bit originally . But it changed to at 48k 32 bit float when music rebalence. Im just not understanding if that was the reason that the key to a song changed from an eb to and A#m on virtual dj
It sounds like both the RX and Protools SRC collapse the stereo image a bit? Protools more than RX. Thanks for doing these vids, the quality of your content has improved massively!
Hmm... I also hear that a bit. It's especially noticeable at 5:09. PT also sounds flatter, or weaker somehow. Not in volume, but in depth. I didn't think I was going to hear any difference at all tbh, but there definitely is some.
My friend sent me an Ableton project that he recorded at 44.1k but I opened it on my end by mistake in 48k and got to working on mixing... did I royally fuck myself? Do I have to start over? It sounds fine but my gut feeling is I blew it 😞
I was just reading the comments and the video was not visible but the moment the original file started paying again I got back to the video as it was such a beautiful sound of the original recording... I am glad I was never tempted/had to do up sampling since I deal with recording the stuff and I choose everything. Also, my gut feeling always was to use 48k when converting to MP3 from 96k rather than 44.1k and something tells me it was (and is) the right thing to do... What are your thoughts?
Depends. 48khz sample rate is good for smart phones and tablets because that is their native sample rate as they are considered media players with mass media storage. 44.1khz is good only if it's going on a streaming platform or a cd and the cd will be inserted into a cd player. Here's the kicker......let's say you sample rate convert your music to 44.1khz so you can send it to steaming services. Now everyone listening to that 44.1khz sampled song will be listening to it on their smart phone or tablet that has a native sample rate of 48khz. A true tragedy.
I have a Z-Systems z-link96+ mini sample rate converter (asynchronous) that I use at live concerts to burn a one-off CD from the output of my 96Khz 24 bit recorder. I think it works well.
DSD is an awesome sounding format but DSD equipment is expensive and nothing is available in DSD which means that you'd be making music for approximately 4 people.
In order to edit DSD you have to convert it to PCM format known as DXD. And you need software that does that called Pyramix. So it's better to just capture using PCM from the beginning.
Looking forward to the demo files. I'm not sure if the differences can really come though on UA-cam but I feel like the Pro Tools loses some of the openness and depth. I also use RX for SRC but I've heard good things about r8brain. There's a really great SRC page on Infinite Wave where you can see the various test like a sweep or an impulse of countless different SRC in programs. It includes all of the ones you mentioned. You can easily see in the graphs how the Pro Tools one is much worse compared to the RX.
HUUUUGE difference.... even on UA-cam - the original being the best... and ProTools the worst. Also, here is the interesting one. I owe UAD. There were only 3 recordings I have done in 192k. Those sound exceptionally warm and natural and although being just small samples of a few songs I am drawn to listen to those more than other recordings. I do not believe it was due to the sample rate as such but because I was using UAD plugins on the way in it means 4 times more processing by UAD than at 44.1k and that was what made it. Unfortunately, I do not get much time to do more experiments as I am in the process of setting up my new studio but I would definitely try more to see if it really works. Would you be able to compare new UAD and the previous version on AD DA to see what is your opinion. Thanks
I get a smoother sound quality upsampling via software before sending my 44.1k files to my DAC. Careful to always use an even multiple; usually 4x. This sounds better because my DAC doesn’t have to use a steep slope (brick wall) filter which will typically introduce phase distortion. Uncertain how this would be different in a mix - although the cost of equipment is much higher the higher the sample rate.
Watching on UA-cam through some decent monitors and I can hear plenty of difference. Stereo image, Top end and depth all compromised on the converted files. The Izotope much less so. Thumbsup.
Need some advice. I am trying to decide between a guitar mfx pedal that records at 48 kHz and another one that records at 44.1 kHz. If I I get the 44.1 kHz one I I might need to upsample for video purposes, and if I get the 44.1 one I might have to downsample for music purposes? Which would you say is better, upsampling or downsampling.
for recording guitar you should record it at the highest frequency possible. If you record at 44,000 you will have no way to auto correct or pitch correct any notes with out losing half your frequency if you record at 44,000 it will sound muddy and really bad.thus 96,000 is good or higher if you can. I made this mistake and I hated the way my guitar sounded for over 20 years until I figured out why. its only when you get the master the way you want it do you render a final out at 44000 or 48000
Peace and blessings to you my brother, I love your channel. Question tho, is there any benefit to upsample from 44-96 while mixing to do processing like time stretching and pitch correcting etc. Will it result in a better sound?
I found the easiest place to hear the difference was from about 7.05 to about 7.35 going from original to the RX (clear difference obviously) and then from RX to Protools.. a more subtle difference.. but a difference still !
That's a good idea if you're using 'non linear' plugins that don't already do oversampling, like some compressors, vintage eqs, distortion, phase effects etc. It should put the aliasing they could cause out of the audible range.
Seems like he's using one interface's Analog stereo OUT to his analog mastering equipment and then monitoring that signal from another interface/computer and that particular monitoring computer is set to the higher sample rate. That way he can actually up-convert the analog processed signal. CLA, when mixing, uses a similar approach. If you have the extra computer (and an extra A/D converter stereo channel) you can do this. If not, just print the final mix/master at the same sample rate.
Doing this on youtube is pointless. I can't hear any difference between them. Please proved the files when you do these tests. I don't mean to call you out Streaky but you should know better. This is like the millions of Morons that do speaker demos and record the sound with a cheap mic, upload it and say... that's my awesome new set up, what do you guys think...Shit, give me a break. I still love you Streaky. I watch your vids because you are spot on most of the time. You would not be sitting in that chair right now if you were not. ... Everybody should read the book 'Why do smart people do stupid things'.
Well, it's true that UA-cam makes it much harder to hear differences, and sounds very different than lossless files, but I still hear difference. What I learned is that it's not possible to judge subtle differences on UA-cam - while differences are heard, conclusions sometimes are drawn completely different from conclusions drawn from lossless files
The biggest reason why you would record at 96,000 or 192,000 would be if you intend manipulate the recording. If you are recording singing or guitar 96,000 will allow you to pitch correct your singer and or your guitar solo to some degree in which it will then cut your frequency in half. You cannot auto tune or pitch correct a 44,000 sample and keep it at 44,000 thus you would no longer have cd quality after the correction. with 192,000 you range of manipulation flexibility and slow down possibility should be even greater. I would encourage anyone to record at the highest frequency possible because at 44,000 you will have no wiggle room to fix or manipulate slowdown harmonize etc etc your work. You should always keep your master recording un manipulated and make a double of it to manipulate it because you may want to revisit your stems it in the future possibly for other work.
The main problem here is that if you don't downsampling to 48/44.1khz your audio will be automatically downsampled and ruined by streaming service that will not let your audio play at the original sample rate, however the max for playback streaming on youtube is 48khz so if your song (original) was higher, we will never know :(
UA-cam resamples everything to 44.1khz. Read their specifications. The tragedy here is that most of the people who use youtube or a music streaming service (44.1khz) will use their smart phone or tablet to do so which have a native internal sample rate of 48khz as it's a media device. UA-cam is a video platform and should have 48khz as their sample rate as that is the video standard sample rate. But this world is ass backwards and youtube sticks to 44.1khz for whatever idiotic reason. So when the blue ray disc of a movie comes out it will be it's original video sample rate of 48khz, but once uploaded to UA-cam it will become 44.1khz and the device you watch it on will have a native sample rate of 48khz.
Exactly the answer I was looking for. Best way to SRC (downsample) at the end of the project. Thanks!
I tend to use the Saracon though I agree it does have a sound to it. RX tends to just not have a sound at all whereas the saracon has an emphasis of sorts around the nyquist
Wondering if I'm following correctly: are you doing this process before or after you've mastered? And wondering how you go about capturing the file to the desired sample rate (the process you mention at 2:00)?
Really good video on this topic! My preference is to pick a relatively high sample rate and bit depth and stick to it as much as possible. If I have to convert the sample rate, I try to do it by a nice clean integer factor. This is usually going to create two common families of sample rates, CD-based, which is 44.1, 88.2, 176.4, 352.8, 705.6, 1411.2, etc… and DVD-based, which is 48, 96, 192, 384, 768, 1536, etc… My preference is to pick a relatively high sample rate from one of these two families and stick to it. If I have to convert it, I try to stay within the same family so that it can go up or down by doubling or halving. When converting, you use a brick wall filter to cut off frequencies above the Nyquist limit of the final sample rate, which is one half of the final sample rate. You then cut out samples at regular intervals using decimation. For example to go from 96 to 48 you only use every other sample since you are going down by a factor of two. 192 to 48 uses every 1st out of 4 samples and skips three for each sample cycle. If I have to convert by a non-integer factor, 48 to 44.1 for example, I just try to use the best interpolation algorithm that I can find to get the audio as close to the original analog continuous waveform signal as possible. Obviously if you need to convert something like bit-depth/sample-size/word-length you round it to the nearest integer. If you need to convert the modulation and/or demodulation method you use an algorithm to get as close as you can. Obviously like you mentioned different algorithms, software, firmware, hardware, tools, AI models, artificial neural-networks and/or equipment can yield better or worse quality results, so it’s definitely a matter of finding a good one like Isotope for example.
Please! What's the name of the song example in the video? Love the comp
RX conversion is better than Protools , thank you Streaky for testing all the stuff and posting it on YT ..
On some of my projects i use samples, which most of 'em are 44k 24bit....but i like to record my hardware synths at 96k 32 floating point. If the arrangement has most of the tracks recorded from hardware synths, is it wrong to upsample those loop samples coming from libraries at 96? Or record the hardware synths at 44k as the lowest sample rate in the mix?!?
That's what I also want to know. ROMplers like Omnisphere and Nexus are 44.1 as well... And I use them a lot...
Should I bounce them as audio track and upsample...
I know things have improved since 44.1/16 times but I found back then that when a waveform is processed too many times it develops a peculiar "sheen" and loses presence. The "generations" problem of analogue tape seemed to have gone away but that didn't mean we could apply more EQ to fix the EQ that we'd applied to fix the EQ. Obviously at todays resolutions that's less obvious but it still makes me think it's wise to avoid unnecessary math.
Thank you so much for the video! May I ask? Is thiis out of date now in terms of SRC as I see logic, ableton and cubase have upgraded their SRC and charts show there to be no issues now?
This is so intriguing. I'll definitely be testing out streaky for a master soon. I love the detail he's taking into trying to get the absolute best sound possible.
I have done so already - he is very good - not just my opinion but everyone's who listed to his master.
I'm listening on headphones (Focal Listen Pro's) and I can actually hear it! The original sounds better...the depth and clarity. I didn't think I would be able to hear it on UA-cam but I can...the silk is in the original but for me I can't hear that in both of the conversions. It's super subtle but there's definitely a difference if focusing on the detail. Like the F1 drivers making changes to win the race with milliseconds...mastering is the equivalent...so in this context I can hear it and it does make a difference. It's a shame that it needs to be down sampled...it definitely sounds the best in it's original form.
I isolated drums and bass of a track. The track is 44100 k 16bit originally . But it changed to at 48k 32 bit float when music rebalence. Im just not understanding if that was the reason that the key to a song changed from an eb to and A#m on virtual dj
what are your thoughts on using voxengo R8Brain for converting. it seem pretty sweet.
It sounds like both the RX and Protools SRC collapse the stereo image a bit? Protools more than RX. Thanks for doing these vids, the quality of your content has improved massively!
Hmm... I also hear that a bit. It's especially noticeable at 5:09. PT also sounds flatter, or weaker somehow. Not in volume, but in depth. I didn't think I was going to hear any difference at all tbh, but there definitely is some.
My friend sent me an Ableton project that he recorded at 44.1k but I opened it on my end by mistake in 48k and got to working on mixing... did I royally fuck myself? Do I have to start over? It sounds fine but my gut feeling is I blew it 😞
I was just reading the comments and the video was not visible but the moment the original file started paying again I got back to the video as it was such a beautiful sound of the original recording... I am glad I was never tempted/had to do up sampling since I deal with recording the stuff and I choose everything. Also, my gut feeling always was to use 48k when converting to MP3 from 96k rather than 44.1k and something tells me it was (and is) the right thing to do... What are your thoughts?
Depends. 48khz sample rate is good for smart phones and tablets because that is their native sample rate as they are considered media players with mass media storage. 44.1khz is good only if it's going on a streaming platform or a cd and the cd will be inserted into a cd player. Here's the kicker......let's say you sample rate convert your music to 44.1khz so you can send it to steaming services. Now everyone listening to that 44.1khz sampled song will be listening to it on their smart phone or tablet that has a native sample rate of 48khz. A true tragedy.
Even on my phone i can hear RX sounds more open
I have a Z-Systems z-link96+ mini sample rate converter (asynchronous) that I use at live concerts to burn a one-off CD from the output of my 96Khz 24 bit recorder. I think it works well.
You're just taking up more space on your hard drive by going above 48Khz, unless you plan to pitch down your samples more than 2 octaves.
he's downsampling it back
Should have been a blind test!
Do you have any experience with DSD? In particular I’m interested in more on DSD conversion to PCM. Thanks
DSD is an awesome sounding format but DSD equipment is expensive and nothing is available in DSD which means that you'd be making music for approximately 4 people.
In order to edit DSD you have to convert it to PCM format known as DXD. And you need software that does that called Pyramix. So it's better to just capture using PCM from the beginning.
hope all is well
Looking forward to the demo files. I'm not sure if the differences can really come though on UA-cam but I feel like the Pro Tools loses some of the openness and depth. I also use RX for SRC but I've heard good things about r8brain. There's a really great SRC page on Infinite Wave where you can see the various test like a sweep or an impulse of countless different SRC in programs. It includes all of the ones you mentioned. You can easily see in the graphs how the Pro Tools one is much worse compared to the RX.
HUUUUGE difference.... even on UA-cam - the original being the best... and ProTools the worst.
Also, here is the interesting one. I owe UAD. There were only 3 recordings I have done in 192k. Those sound exceptionally warm and natural and although being just small samples of a few songs I am drawn to listen to those more than other recordings. I do not believe it was due to the sample rate as such but because I was using UAD plugins on the way in it means 4 times more processing by UAD than at 44.1k and that was what made it. Unfortunately, I do not get much time to do more experiments as I am in the process of setting up my new studio but I would definitely try more to see if it really works. Would you be able to compare new UAD and the previous version on AD DA to see what is your opinion. Thanks
What settings do you use in isotope?
I get a smoother sound quality upsampling via software before sending my 44.1k files to my DAC. Careful to always use an even multiple; usually 4x. This sounds better because my DAC doesn’t have to use a steep slope (brick wall) filter which will typically introduce phase distortion. Uncertain how this would be different in a mix - although the cost of equipment is much higher the higher the sample rate.
Watching on UA-cam through some decent monitors and I can hear plenty of difference. Stereo image, Top end and depth all compromised on the converted files. The Izotope much less so. Thumbsup.
Need some advice. I am trying to decide between a guitar mfx pedal that records at 48 kHz and another one that records at 44.1 kHz. If I I get the 44.1 kHz one I I might need to upsample for video purposes, and if I get the 44.1 one I might have to downsample for music purposes? Which would you say is better, upsampling or downsampling.
for recording guitar you should record it at the highest frequency possible. If you record at 44,000 you will have no way to auto correct or pitch correct any notes with out losing half your frequency if you record at 44,000 it will sound muddy and really bad.thus 96,000 is good or higher if you can. I made this mistake and I hated the way my guitar sounded for over 20 years until I figured out why. its only when you get the master the way you want it do you render a final out at 44000 or 48000
Peace and blessings to you my brother, I love your channel. Question tho, is there any benefit to upsample from 44-96 while mixing to do processing like time stretching and pitch correcting etc. Will it result in a better sound?
No because there is no information recorded at 96khz. It's like taking a digital picture and upscaling it. You aren't adding any new detail to it.
I found the easiest place to hear the difference was from about 7.05 to about 7.35 going from original to the RX (clear difference obviously) and then from RX to Protools.. a more subtle difference.. but a difference still !
Izotope rx indeed sounds great
The best for the money
What would be your thoughts as a mix engineer importing a whole session at 44.1/48kHz and upping that to 96k for mixing?
Works for me , the resolution gets better , plugins sound better at 96k
Ridiculous because then you should just record at 96khz to begin with.
That's a good idea if you're using 'non linear' plugins that don't already do oversampling, like some compressors, vintage eqs, distortion, phase effects etc. It should put the aliasing they could cause out of the audible range.
Seems like he's using one interface's Analog stereo OUT to his analog mastering equipment and then monitoring that signal from another interface/computer and that particular monitoring computer is set to the higher sample rate. That way he can actually up-convert the analog processed signal. CLA, when mixing, uses a similar approach. If you have the extra computer (and an extra A/D converter stereo channel) you can do this. If not, just print the final mix/master at the same sample rate.
Pro Tools conversion sounds hollow.
As with every A/B test ever, Pro Tools sounds the worst.
i still think the original sounds better
Doing this on youtube is pointless. I can't hear any difference between them. Please proved the files when you do these tests. I don't mean to call you out Streaky but you should know better. This is like the millions of Morons that do speaker demos and record the sound with a cheap mic, upload it and say... that's my awesome new set up, what do you guys think...Shit, give me a break. I still love you Streaky. I watch your vids because you are spot on most of the time. You would not be sitting in that chair right now if you were not. ... Everybody should read the book 'Why do smart people do stupid things'.
he mentions that the files will be uploaded so maybe have a little patience?
They will be in a link below just waiting for Dropbox to update :(
Check 03:04 and have a little patience.
Ok, you got me.
Well, it's true that UA-cam makes it much harder to hear differences, and sounds very different than lossless files, but I still hear difference. What I learned is that it's not possible to judge subtle differences on UA-cam - while differences are heard, conclusions sometimes are drawn completely different from conclusions drawn from lossless files
first