Blog on TuneTrainer ​

I'm a little concerned about the viability of TuneTrainer.

As someone who's pretty heavily focused into machine learning(sadly not NLP, but my feedback may still be valuable? idk) and do music as a hobby, I have some doubts GPT-3.5 or 4 can really generate song lyrics with proper syllables in a way that can be simple enough to be used for studying.

All the songs that I know for common topics during elementary/middle school were very simple in nature, and it's pretty hard to ask GPT to come up with an actually viable song, when it can't really come up with any musical beats in the first place- a weakness that is rarely discussed.

Even if GPT could come up with syllables, I think it'd have a hard time with rhythm, which is a very important part of a song.

To make it actually viable, you could make a list of songs and keep track of their rhythm and syllables, simple ones that could be used for lots of different pieces. You could then perform syllable analysis on the lyrics you actually want and overlay it over the existing pieces.

This is not an up-to-scale solution, but it seems like it could work for now.