search

SonyCSLParis / pesto

Self-supervised learning for fast pitch estimation
GNU Lesser General Public License v3.0
168 stars 15 forks source link

Questions about online processing #30

Closed kmizuki closed 1 month ago

kmizuki commented 2 months ago

Do you have any ideas on how to run pesto processing online? Also, what is the default model's training dataset?

Thank you.

aRI0U commented 1 month ago

Hi! By online you mean processing a stream of audio frame by frame in real-time? The model itself is extremely lightweight and processes CQT frames independently so it should be fast enough, however computing the CQT in real-time can be challenging because of the large kernel required for low frequencies. I'm currently not aware of a solution for solving this issue.

About the training set the provided checkpoint has been trained on MIR-1K.