This release introduces Vulkan GPU support for whisper.cpp making it significantly faster even on laptops. Real-time transcription is possible even with large models on computers with ~5GB RAM video cards. There is now an option to separate voice tracks before the audio is transcribed. This can improve transcript accuracy for audios with background noises or music. Faster whisper was updated to the latest version adding noticeable speed improvement.
Additional improvements
- Option to switch the UI language from preferences
- Library updates for better Linux compatibility, especially in Flatpak installations
- Option to upload live transcripts to a server. See BUZZ_UPLOAD_URL
- Search and additional controls in Transcription viewer by @shlomi-dr
- Added UI translation for German, Dutch, Danish and Portuguese (Brazilian)
- Minor bug fixes
New Contributors
- @oleguldberg made their first contribution in #1045
- @Ironship made their first contribution in #1120
- @Vistaus made their first contribution in #1123
- @Elandrya2711 made their first contribution in #1156
- @GoldMath made their first contribution in #1158
- @ericdq made their first contribution in #1168
- @prschopf made their first contribution in #1190
- @shlomi-dr made their first contribution in #1224
- @greatdaveo made their first contribution in #1246
Full Changelog: v1.2.0...v1.3.2