Skip to content

Changelog

1.0.0

New features:

  • Job queue: while a job runs, Add to Queue prepares the next ones, which then run in order. The job list shows each job's state, with actions on right-click.
  • Playlists: a playlist page's videos are all split, one after the other.
  • Links to the written stems in the status log, and an Open Output Folder menu entry.
  • Progress is also shown while downloading the video and the model.
  • More YouTube URLs are recognized: youtu.be, Shorts, live videos, m.youtube.com and YouTube Music.
  • --version option.
  • Stems are written at 48 kHz by default, the usual rate of DAW sessions, which then don't have to convert them on import. The models still work at 44.1 kHz; the stems are resampled afterwards. 44.1 kHz can be chosen in Settings..., or for one run with --sample-rate.

Fixes:

  • Stop now stops the job immediately and cleanly, at any step, instead of abruptly killing the thread: the separation runs in a child process, which Stop ends, and the job's intermediate files are removed. An interrupted model download no longer leaves a truncated model behind. The button is red.
  • GPU acceleration on Windows: the GPU environment got PyPI's CPU-only PyTorch. It now gets PyTorch's CUDA build (CUDA 13 for Blackwell GPUs).
  • OpenUnmix produces stereo stems (they were mono), and is given the audio's actual sample rate.
  • The audio is decoded straight to FLAC, without the lossy MP3 step, which degraded the separation's input.
  • A failed step no longer reports "Extraction complete", and errors are reported once, with their cause.
  • Closing Yaas during a job no longer crashes it.

Other changes:

  • Removed the unused ffmpeg PyPI dependency (Yaas uses the ffmpeg program).
  • Test suite, run by CI.

Yaas is versioned with bumpver and released via tagged GitHub Releases.

See the GitHub Releases page for the full history of published versions and their artifacts, and the commit history for day-to-day changes.