Thanks, that's pretty hands-on. Gradint looks like a pretty old program, and I couldn't quite get from its page what it is that it does exactly. Does it simply take a collection of phrases recorded as separate audiofiles and then combines them lessons using a spaced repetition algorithm? It doesn't actually parse those individual files in any way, right? For instance, it won't try to break a phrase into words and words into syllables. Also I was not sure at which point in the presentation you ran Gradint exactly. Was it triggered directly from your Python script / program?
It doesn't break the phrase up or do reverse pronunciation like Pimsluer. You'd need a human for those sorts of things. It doesn't parse the files, it just takes the collection of files and applies the Pimsluer algorithm to them for space repetition.
@@Rick62820 thanks for the response. I think parsing on syllables and/or word boundaries is very doable, but it’ll obviously require a library / pretrained neural network to avoid reinventing the wheel. Now I have a side project in mind, haha
@@Rick62820 well, text tokenizers have existed since a long time ago. Surely with the recent breakthroughs in speech recognition, there should now be speech tokenizers too. Worst case scenario, one would need to train a neural network on one’s language of choice. And if it very occasionally breaks up words incorrectly, it probably isn’t such a big deal after all. And then you can make it accept actual dialogs / audio files created for learners of that language from some traditional courses, break them up into phrases and phrases into segments (words and syllables) and then leverage spaced repetition. Effectively, you would be able to generate an audio course identical to Pimsleur from a bunch of dialogs / recordings
a very interesting way to create audio files but this is too technical for me...😅
You should try, it isn't as bad as it seems.
Thanks, that's pretty hands-on. Gradint looks like a pretty old program, and I couldn't quite get from its page what it is that it does exactly. Does it simply take a collection of phrases recorded as separate audiofiles and then combines them lessons using a spaced repetition algorithm? It doesn't actually parse those individual files in any way, right? For instance, it won't try to break a phrase into words and words into syllables. Also I was not sure at which point in the presentation you ran Gradint exactly. Was it triggered directly from your Python script / program?
It doesn't break the phrase up or do reverse pronunciation like Pimsluer. You'd need a human for those sorts of things. It doesn't parse the files, it just takes the collection of files and applies the Pimsluer algorithm to them for space repetition.
I ran the gradint python script by itself. The other script was just to get the audio files.
@@Rick62820 thanks for the response. I think parsing on syllables and/or word boundaries is very doable, but it’ll obviously require a library / pretrained neural network to avoid reinventing the wheel. Now I have a side project in mind, haha
Interesting! If do it I am sure the gradint author would be interested. I certainly would be.
@@Rick62820 well, text tokenizers have existed since a long time ago. Surely with the recent breakthroughs in speech recognition, there should now be speech tokenizers too. Worst case scenario, one would need to train a neural network on one’s language of choice. And if it very occasionally breaks up words incorrectly, it probably isn’t such a big deal after all. And then you can make it accept actual dialogs / audio files created for learners of that language from some traditional courses, break them up into phrases and phrases into segments (words and syllables) and then leverage spaced repetition. Effectively, you would be able to generate an audio course identical to Pimsleur from a bunch of dialogs / recordings