I am working on an app for rhythm training for dancing.
The first step is beat detection in music - here is my code for generating videos with beat counting: https://github.com/zby/beat_counter (vibe coding warning). Beat detection seems to work OK, downbeats are a bit more tricky and I have found not good solutions for further structuring of music (for dancing they usually count to 8 - that is two 4 beat measures - or 6 - two 3 beat measures for Waltz for example, I am not sure about 2/2 measures and other). Seeing all the LLMs generating music I thought there should also be LLMs interpreting music - but so far I have found no such model the available algos seem to be from a previous decade.
My current plan is to test the counting with people with good rhythm sense and once I find a good algo for beat detection I'll proceed with writing the app.
I assume that it would mostly help beginners (and maybe instructors) - but we'll see. If you send me an audio file I can generate the counting video (with the current model) - I am curious of your opinion.
My current plan is to test the counting with people with good rhythm sense and once I find a good algo for beat detection I'll proceed with writing the app.