AttributeError: `np.NaN` was removed in the NumPy 2.0 release. Use `np.nan` instead.

[rhdunn]

I'm getting this error when running dataspeech's main.py with python 3.12.11 and the packages installed into a virtual environment.

It is originating in pyannote/audio/core/inference.py line 529. According to pyannote/pyannote-audio#1758 version 3.3.2 has a fix.

If I pass pyannote.audio==3.3.2 to the pip command installing requirements.txt I get an error that "brouhaha 0.9.0 depends on pyannote.audio==3.2.0". The brouhaha library is being installed from source and has a fixed version dependency.

The workaround is to force the version of numpy via:

pip install -r requirements.txt numpy~=1.0

Adding the following line to requirements.txt resolves the issue by doing the same thing -- i.e. limiting numpy to 1.x:

numpy~=1.0

When I run the pip command that installs numpy 1.26.4 instead of the incompatible 2.2.6.

Package Versions:

• numpy == 2.2.6
• pyannote.audio == 3.2.0
• brouhaha == 0.9.0 commit 697f6a35998c1edcf557169ba16fa930def04e7a

[alunkingusw]

I have the same problem when importing This issue needs to be raised on the brouhaha github page, so I have added it there.
As a temporary fix, I forked brouhaha and altered the requirements.txt to pyannote.audio = 3.3.2 and now it installs for me OK. You can fork it yourself (recommended) or manually install https://github.com/alunkingusw/brouhaha-vad/archive/main.zip

[alunkingusw]

The numpy version has been added to brouhaha to enforce numpy version 1. It would be interesting to see if this works for this project?

[rhdunn]

As noted in my original post, limiting numpy to version 1 (numpy~=1.0) resolved this issue for me.

[alunkingusw]

Awesome! Thanks 👍

[alunkingusw]

Since it is fixed at source on brouhaha-vad, if there are no errors, this issue can be closed.