Closed lostanlen closed 4 years ago
As noted on #3, we should pin the version of librosa to whichever version we used to compute the features that were used to train the model we're distributing. If we want to use a new release (or any release that's not the one used to compute the features used to train the models previously), we'll have to recompute the features and retrain the model.
After giving it more thought today, i don't think it would necessarily simplify the process. Let's not fix what isn't broken.
Closing.
https://github.com/librosa/librosa/issues/772 (closed by https://github.com/librosa/librosa/pull/872) allows streaming in librosa 0.7.0. This means that we could simplify block streaming in
process_file
.Furthermore, we should propagate the latent variable (
zi
) inPCEN
now that this is a return value inlibrosa.pcen
. (https://github.com/librosa/librosa/pull/864, closed https://github.com/librosa/librosa/pull/816 and https://github.com/librosa/librosa/pull/851).