Closed carlosholivan closed 4 years ago
CREPE was designed for monophonic pitch transcription. Changing the model architecture and re-training would basically amount to developing a different model. In short, CREPE is not designed for polyphonic transcription.
@carlosholivan for polyphonic transcriptions i suggest you to look at this project. The results and demo seems to be quite promising
@carlosholivan Another good option for polyphonic transcription is Deep Salience: https://github.com/rabitt/ismir2017-deepsalience.
It's a couple of years old now but was state-of-the-art when released and I believe it's still pretty competitive (and easy to use).
Closing this out, cheers
I was wondering if Crepe model would allow transcriptions of instruments with polyphonic notes (for example chords played by a violin), maybe changing somehow the model and retraining it would allow it to handle polyphony?