What's New
- Added optional MMS forced alignment to refine word-level timestamps after transcription (enable in settings).
- This will be a huge improvement for models like Cohere and Canary, as these models do not output any timestamps natively - just raw text - so I had to interpolate them based on the audio chunk which leads to relatively bad timing data.
Bug Fixes
- Fixed Resolve audio export incorrectly reporting failure when DaVinci Resolve is set to a non-English locale.
Improvements
- Improved settings dialog readability with a scroll-down hint and better forced-alignment placement.