Built a local macOS menubar dictation app on top of VibeVoice-ASR
Hey everyone huge fan of this model.
I built VibeDictate, a lightweight macOS menubar app that wraps VibeVoice-ASR into a daily-use dictation tool. Hold
⌥Space to record, release to transcribe; it pastes directly into whatever app you're in.
Why I built it: Every cloud dictation tool either sends your audio somewhere, costs per minute, or requires internet.
VibeVoice-ASR is the first model I've found that's clean enough to use as a real daily driver so I wrapped it into
something I'd actually keep running all day.
A few things I added on top:
- Auto-unloads the model after 5 min of inactivity to reclaim the ~28GB RAM
- Strips [Noise] / [Environmental Sounds] tags from output automatically
- History menu — last 5 transcriptions, click to copy
- One-command install script + optional launch-at-login via launchd
Repo: https://github.com/Sohaibsajid50/vibedictate
Would love feedback from anyone using the model. Also looking for collaborators on:
- MPS (Metal GPU) support — currently forced to CPU due to a Metal driver assertion on the float32 weights
- Streaming / partial transcription output
- .app bundle packaging so non-technical users can install it
If you've gotten this model running on MPS, please drop a comment that's the one thing I haven't cracked yet.