debpalash/OmniVoice-Studio v0.3.5

OmniVoice Studio v0.3.5

on GitHub

Fixed

• Speaker diarization failed on PyTorch ≥ 2.6 (Weights only load failed … Unsupported global: torch.torch_version.TorchVersion) even with the pyannote
  license accepted. PyTorch 2.6 made torch.load default to
  weights_only=True, whose secure unpickler rejects the pyannote checkpoint's
  metadata globals. The diarization loader now registers the same safe-globals
  allowlist the WhisperX VAD load already uses, so the secure load succeeds.
  (#270)

Linux x64 artifacts

64e81fa1f8900c883dfa8ce700abe6f7a9de41e26013e5f5e06d8014bbbe2fc4  OmniVoice Studio_0.3.5_amd64.AppImage
00cf608105b0c4fb25122f60bdcf39382c7b6459305ce91b74bc0d1b30bb673d  OmniVoice Studio_0.3.5_amd64.AppImage.sig

Windows x64 artifacts

39529182a1c61a561fffc82c82c5886b8e4c640bf153c6134d9570099192e906 *OmniVoice Studio_0.3.5_x64_en-US.msi
462243782917d945ddd2c9b463da3e7a094e18205a36bdd3be02d92d6b4c893c *OmniVoice Studio_0.3.5_x64_en-US.msi.sig