Closed mweinelt closed 1 year ago
Weird but then the CI would fail. Is there something I am missing?
Numba still constrains the numpy version to below 1.24.0, see https://github.com/numba/numba/blob/288a38bbd5a15418a211bf067878dfdf3c139509/setup.py#L26, that's why this hasn't become an issue on the CI.
We (nixpkgs) backported numpy 1.24 support for numba, that's why we're running into this error. I'm currently overriding the dependencies for TTS
So my ask is probably about updating librosa to a newer release with numpy 1.24 support. The change required seems to be available from 0.9: https://github.com/librosa/librosa/commit/a9598f9d5a68186c14499f614e2b234fc932fdef
This might help #2480 ?
Thanks, it will!
This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions. You might also look our discussion channels.
Numpy is now indeed relaxed, but the librosa version pinned is incompatible with latest numpy(==1.24.2) version. In fact tts was previously already working against numpy 1.23.5, and that is still the latest version that works for me after this release.
Importing librosa on an environment with numpy 1.24 leads to the following traceback for me. This can happen because librosa only sets a lower bound for its numpy version.
https://github.com/librosa/librosa/blame/0.8.0/librosa/core/constantq.py#L1058
The
np.complex
attribute was dropped in numpy 1.24.0.https://numpy.org/devdocs/release/1.24.0-notes.html
Originally posted in https://github.com/coqui-ai/TTS/issues/2315#issuecomment-1474333266, but reposting here to make it actionable.