After being slammed with calls for the past three weeks, I'm finally back in a more stable developer cadence! This release adds NVIDIA's Parakeet model alongside our existing Whisper support, and several accumulating stability issues (more fixes to come). Thank you to everyone who's been pushing issues and pull requests during the hiatus; I hope to review them shortly.
The big change: I migrated from whisper-rs to transcribe-rs,, which fixes the GPU acceleration issues many of you reported and lets us support multiple transcription engines. Huge thanks to @cjpais for creating this library, we're proud to sponsor you! 🫶 This was the most important thing to push out. Next up is stabilizing Linux support, followed by more features and fixes.
NVIDIA Parakeet Support
Whispering now supports NVIDIA NeMo's Parakeet ASR model alongside Whisper. For many people, Parakeet is both faster and more accurate than the Whisper.cpp models—it's been called "magic" by users who've tried it.
The Parakeet TDT 0.6B v3 model is ~670MB. It uses GPU acceleration when available and runs transcription locally. It's still new, so expect some more optimizations and fixes to come.
The model selector now handles both Whisper and Parakeet models from one interface. This works because we migrated from whisper-rs to transcribe-rs, giving us a consistent API for multiple transcription engines and also fixing the GPU acceleration issues that caused slower performance for some users.
It is hard to overstate just how complicated this was to set up before, especially with the number of weird edge cases that kept coming up from whisper-rs (don't get me started on the feature flags in 15.0.0 vs 13.2.0). Thankfully, we now have the base required to deploy more fixes and features in the future.
Stability Fixes
This release fixes a few smaller issues that have been accumulating over the past few months:
- FFmpeg compatibility: Better detection and clearer warnings when FFmpeg is missing
- macOS DMG bundling: Fixed CI/CD issues that were preventing DMG creation
- Error messages: More descriptive errors when things go wrong (especially for transcription failures)
- UI fixes: Notification button styling, accessibility flows, active state highlighting
- Linux AppImage: Added X11 dependencies and removed Wayland to fix compatibility issues
I've also done some housekeeping on the codebase.
What's Changed
- refactor: migrate oninput handlers to function bindings and bind:value by @braden-w in #763
- refactor(path-utils): improve path handling and move compression logic by @braden-w in #764
- fix: update workflows to use PAT instead of GITHUB_TOKEN by @braden-w in #765
- chore(whispering): remove deprecated groq model by @nocdn in #759
- fix(EditRecordingModal): prevent audio rerendered by @thisisharsh7 in #734
- Added reference docs - batch 1 by @onel in #783
- Fix(docs): point docs to our docs and server by @rupokghosh in #787
- feat: migrate to transcribe-rs for Local Transcription by @braden-w in #791
- feat: add Parakeet model support for fast local transcription by @braden-w in #800
- refactor(transcription): reorganize services by deployment type by @braden-w in #801
- refactor: improve FFmpeg compatibility checks with dependency injection by @braden-w in #804
- chore: remove eslint perfectionist plugin and sorting rules by @braden-w in #803
- build(ci): X11 deps and strip Wayland from AppImage by @braden-w in #805
- refactor(config): remove perfectionist ESLint plugin by @braden-w in #806
- Docs: Update AGENTS.md to prevent unexpected destructive/expensive tool calls. by @Leftium in #802
- docs: README for icons directory by @braden-w in #807
- feat(transcription): overhaul local model download system by @braden-w in #808
Acknowledgments
Special thanks to:
- @cjpais for creating the excellent
transcribe-rslibrary and Handy ❤️. Handy is an awesome cross-platform desktop speech-to-text application that I personally use, especially when developing Whispering locally. - @ggerganov for whisper.cpp, the foundation of local Whisper transcription
- NVIDIA for the Parakeet ASR models
- All our contributors who helped make this release possible
New Contributors
Full Changelog: v7.4.1...v7.5.0
Questions? Join our Discord or check the README.
Love Whispering? Star us on GitHub to show your support!