Skip to content

feat: add Moonshine Base speech recognition model#556

Merged
cjpais merged 3 commits intocjpais:mainfrom
vladstudio:feat/moonshine-support
Jan 10, 2026
Merged

feat: add Moonshine Base speech recognition model#556
cjpais merged 3 commits intocjpais:mainfrom
vladstudio:feat/moonshine-support

Conversation

@vladstudio
Copy link
Contributor

@vladstudio vladstudio commented Jan 8, 2026

(human):

hey! long time no see. I thought I'd add another model to the list. More choices always better!

(llm):

Add support for the Moonshine Base ASR model, which offers:

  • 5x faster transcription than Whisper
  • Better accent recognition
  • ~240MB model size (encoder + decoder + tokenizer)

Changes:

  • Update transcribe-rs to v0.2 with moonshine feature
  • Add EngineType::Moonshine variant
  • Add Moonshine Base model definition (400MB, 80% accuracy, 90% speed)
  • Add model loading and transcription logic
  • Add i18n translation for Moonshine Base

Note: Model files must be hosted at blob.handy.computer/moonshine-base.tar.gz
Expected structure: encoder_model.onnx, decoder_model_merged.onnx, tokenizer.json

Co-Authored-By: Claude Opus 4.5 noreply@anthropic.com

vladstudio and others added 3 commits January 8, 2026 18:43
Add support for the Moonshine Base ASR model, which offers:
- 5x faster transcription than Whisper
- Better accent recognition
- ~240MB model size (encoder + decoder + tokenizer)

Changes:
- Update transcribe-rs to v0.2 with moonshine feature
- Add EngineType::Moonshine variant
- Add Moonshine Base model definition (400MB, 80% accuracy, 90% speed)
- Add model loading and transcription logic
- Add i18n translation for Moonshine Base

Note: Model files must be hosted at blob.handy.computer/moonshine-base.tar.gz
Expected structure: encoder_model.onnx, decoder_model_merged.onnx, tokenizer.json

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
@cjpais
Copy link
Owner

cjpais commented Jan 10, 2026

For now I pulled in the quantized version and put that up on the handy blob. May change to full model depending on how well moonshine performs for everyone.

@cjpais cjpais merged commit 1b580d5 into cjpais:main Jan 10, 2026
2 checks passed
h0lybyte added a commit to KBVE/Handy that referenced this pull request Jan 12, 2026
…, and UI improvements

Merged upstream features:
- feat: add Czech translation (cjpais#568)
- fix: keybinding changes failing silently due to incorrect key ordering (cjpais#524)
- fix: improve apple intelligence ui and add reusable alert component (cjpais#517)
- Fix race condition when toggling transcription via SIGUSR2 (cjpais#560)
- feat(linux): add ydotool support for text input (cjpais#557)
- feat: add Moonshine Base speech recognition model (cjpais#556)
- Prevent highlight and selection cursor hover on UI text items (cjpais#541)
- feat(ui): add reusable Tooltip component and integrate with settings (cjpais#538)
- PR guidelines for AI
- dont allow package-lock with a bun project for now

Conflicts resolved:
- src-tauri/src/shortcut.rs: Kept KBVE filler detection commands, adopted upstream's improved doc comment
- src/bindings.ts: Merged KBVE types (Discord, Memory, Filler) with upstream's Moonshine engine type
aidynamicsolutions pushed a commit to aidynamicsolutions/codictate that referenced this pull request Feb 21, 2026
Brings in latest main changes since PR cjpais#473 including:
- cjpais#524: keybinding ordering fix
- cjpais#517: Apple Intelligence UI improvements with Alert component
- cjpais#560: SIGUSR2 race condition fix
- cjpais#557: ydotool support for Linux
- cjpais#556: Moonshine Base model support
- cjpais#541: UI text selection prevention
- cjpais#538: Tooltip component
- cjpais#540: Portuguese language support
- cjpais#536: Ukrainian language support

Conflict resolutions:
- PostProcessingSettings.tsx: Combined MLX provider support with new Alert UI
- actions.rs: Fixed async block syntax issue
- Lock files: Regenerated
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants