desktop_v1.4.26 v6.2.3
medium risk
minitap-ai/mobile-useSep 23, 2026 → Sep 23, 20260 days between releases
What's New
Torchaudio dependency is now optional, requiring explicit backend installation via extras.
Breaking Changes
- Default installation no longer includes torchaudio.
- read_audio and save_audio helpers now raise errors if no audio backend is installed.
New Features
- Support for multiple audio backends including torchaudio and torchcodec.
- Lazy import of sequence path to prevent import errors.
- NumPy is now available via [onnx-cpu] and [onnx-gpu] extras.
Improvements
- Reduced base installation footprint by making torchaudio optional.
- Flexible audio I/O handling with multiple supported backends.
Bug Fixes
- Fixed import failure when onnxruntime is not installed.
- read_audio and save_audio now provide clear error messages if backends are missing.
Upgrade Notes
If you use audio helpers, update your install command to `pip install silero-vad[audio]` or `silero-vad[codec]`. If you load audio manually as a torch.Tensor, no changes are required.
The Two Releases
v6.2.3
v6.2.3 Make torchaudio optional
Sep 23, 2026· 29 days ago
Torchaudio is now an optional dependency, requiring users to install specific backends like torchaudio or torchcodec for audio I/O helpers.
Full release notes