search

dodiku / AudioOwl

Fast and simple music and audio analysis using RNN in Python 🕵️‍♀️ 🥁
MIT License
281 stars 21 forks source link

So-so detection ? #8

Open ksuedile opened 4 years ago

ksuedile commented 4 years ago

Hello, i tried to run your lib against the carmen overture from bizet, and it resulted as only 32 beats found...

Humphreybas commented 3 years ago

I also couldn't get really satisfying results so I looked in the code and realised that for good results you should dive into the modules mentioned in the readme:

AudioOwl is using librosa and RNN models to run fast analysis of music files

This AudioOwl module gives a very simple interface to these modules, I think we should rather use it as an example how to use librosa and RNN models than as a fits-all solution to do beat analysis.

@dodiku , thanks for your work. I think it might be helpful to add a little note to the readme, mentioning that for improved detection people should dive into librosa/RNN models.

dodiku commented 3 years ago

@ksuedile Thanks for the feedback. Beat detection is still a challenging task for small Python packages, but I'm sure there'e room for improvement here. Since the last update of AudioOwl, librosa has improved and new RNN and CNN models have been proven to be more accurate for some cases. A more robust interface for AudioOwl might be able to provide these options.

@Humphreybas That's a good idea. I'm not sure when I'll be available to update AudioOwl, but I'd appreciate any pull request.

I appreciate both of your feedback!