4.0.2

junegunn/fzf4.0.2Nov 19, 2025by github-actions[bot]

AI Summary

This patch release addresses a critical segmentation fault by pinning torch versions. It introduces a new `Pipeline.cuda()` convenience method, a preload option for audio, and a CLI option to apply pipelines on directories. It also changes the return type of `Binarize.__call__` from integers to strings.

Key Highlights

Fixed segmentation faults by pinning torch, torchcodec, and torchaudio versions.
Added `Pipeline.cuda()` convenience method for GPU execution.
Added `preload` option to force preloading audio in memory.
Added CLI option to apply pipeline on a directory of audio files.
Improved `permutate` function speed.

Breaking Changes

Made `Binarize.__call__` return `string` tracks instead of `int`.

New Features

Pipeline.cuda() method
Preload option for audio
CLI directory application
Faster permutate function

Full Release Notes

## What's Changed

- BREAKING(util): make `Binarize.__call__` return `string` tracks (instead of `int`) [@benniekiss](https://github.com/benniekiss/)
- fix(torch): pin `torch`, `torchcodec`, and `torchaudio` versions to [avoid segmentation fault](https://github.com/meta-pytorch/torchcodec/issues/995) 
- fix(pyannoteAI): update pyannoteAI wrapper to return both regular and exclusive diarization
- feat(pipeline): add `Pipeline.cuda()` convenience method [@tkanarsky](https://github.com/tkanarsky/)
- feat(pipeline): add `preload` option to base `Pipeline.__call__` to force preloading audio in memory ([@antoinelaurent](https://github.com/antoinelaurent/))
- feat(cli): add option to apply pipeline on a directory of audio files
- improve(util): make `permutate` faster thanks to vectorized cost function

## New Contributors
* @tkanarsky made their first contribution in https://github.com/pyannote/pyannote-audio/pull/1950

**Full Changelog**: https://github.com/pyannote/pyannote-audio/compare/4.0.1...4.0.2