Performance on Instrumental Audio Tracks

marl / crepe

CREPE: A Convolutional REpresentation for Pitch Estimation -- pre-trained model (ICASSP 2018)

https://marl.github.io/crepe/

MIT License

1.13k stars 160 forks source link

Performance on Instrumental Audio Tracks #64

Open chandeepadissanayake opened 4 years ago

chandeepadissanayake commented 4 years ago

Is Crepe basically trained on vocals (speeches)? I tried myself to track pitches on separate vocals and instrumental tracks of the same song and it resulted in considerably high pitch differences at the same time points for vocals and instrumental.

justinsalamon commented 4 years ago

I'm not sure I fully understand the question, but, CREPE was trained on a combination of speech and other instrument sounds, and is designed to work on any pitched sound (speech or otherwise) as long as it's monophonic (i.e. single voice/instrument, no accompaniment). It will not work (as expected) on mixtures of instruments.