QuentinFuxa/WhisperLiveKit 0.2.13

on GitHub

New Features

• --model-path can now point to either a PyTorch file or a folder. Optimized SimulStreaming encoders, mlx-whisper and faster-whisper, are now supported. See Models Compatible Formats.
• Voice Activity Detection is now based on Silero VAD v6. Support for the .onnx format is coming soon.
• The simultaneous translation engine backend has been separated into a dedicated NLLW package, allowing faster development. The current version is 0.1.0, with new releases expected more frequently than WhisperLiveKit updates. WhisperLiveKit 0.1.13 with NLLW 0.1.0 backend performs better than WhisperLiveKit 0.1.12, but is still unstable. Installation using pip install whisperlivekit[translation] now installs nllw backend.
• API changes are still in development. The core is evolving to enable a smoother transition.

Bug Fixes

• Fixed UTF-8 character corruption in the SimulStreaming backend (issue #261).
• Fixed translation issues #257 and #258.

Thank you to all contributors and community members who reported issues and provided valuable feedback.