datasets ~=1.18 requirement introduces conflict w/ numpy > 1.23

[gregoryfoster]

Describe the Bug

speakerbox requires datasets~=1.18 which in turn requires numpy>=1.17. The datasets 1.x line hasn't been updated since mid-2022. numpy v1.20 deprecated use of aliases of built-in types and expired those deprecations in v1.24. The datasets 1.x line uses those aliases, so a default install of speakerbox fails during training.

Expected Behavior

Training should work.

Reproduction

pipenv install speakerbox==1.2.0 Use speakerbox.train()

Environment

• OS Version: macOS 13.4
• Python: 3.10.13
• speakerbox Version: 1.2.0

[gregoryfoster]

To workaround, install numpy <= v1.23.

[evamaxfield]

Is the fix for this just updating the numpy pin / the datasets pin in the library?

[kristopher-smith]

This could probably benefit from a requirements.txt file.