New Features
- Added support for custom models with the simulstreaming backend via the
--model_dirparameter. - Added the ability to set a custom alignment head when using custom models, using
--custom-alignment-heads. If not specified, the system defaults to using the last half of the decoder layers for time alignment. - Unified HTML, CSS, and JavaScript between the Chrome extension and web interface:
- The extension interface now includes the latest functionality.
- The web interface features improved responsive design for narrow viewports.
- Added the
--forwarded-allow-ipsparameter to enable WebSocket functionality behind a reverse proxy. - Prepared backend for the new API version, doc here
Bug Fixes
- Fixed an issue where silences were not detected when the language was not set to auto.
- Fixed a bug causing translation to break during silences.
- Enforced
language = enfor.enmodels. - Frontend: resolved diarization buffer gaps
- Frontend: Speaker, detected language, and lag blocks are now correctly aligned.
- Fixed an issue preventing the frontend from working when
--pcm-inputwas set. - Fixed an issue preventing Diart diarization backend to work. [backend still deprecated]
Thank you to all contributors and community members who reported issues and provided valuable feedback.