Copilot Speech

Private, local voice dictation for GitHub Copilot Chat in desktop VS Code
_{Speak naturally. Review the prompt. Send when you are ready.}

Copilot Speech keeps microphone audio inside an isolated native helper, transcribes it with a local streaming model, and prefills Copilot Chat for review. No cloud transcription service, no automatic submission, and no transcript history.

Highlights

• Powered by Moonshine AI - speech recognition runs entirely on your machine with the local Moonshine Medium Streaming model. Moonshine's benchmarks put Medium Streaming at 6.65% WER on the OpenASR leaderboard — better than Whisper Large v3 (7.44%) with about one-sixth the parameters (245M vs 1.5B) and end-of-phrase latency around 100ms for short utterances instead of multi-second waits. Nothing is ever sent to the cloud.
• Your voice stays private - audio never leaves your device, and no transcript history is kept.
• Responsive as you speak - text appears live while you talk, so you always know it's listening.
• Speak your language - choose English, Arabic, Spanish, Japanese, Korean, Ukrainian, Vietnamese, or Chinese. Non-English Moonshine model weights use the Moonshine Community License for non-commercial use.

Quickstart

Requires desktop VS Code 1.124+ (not the browser).

1. Install Copilot Speech from the Extensions view.
2. Press Ctrl+Alt+V / Cmd+Alt+V (or run Copilot Speech: Start Chat Dictation).
3. Speak. Partial text appears as you talk.
4. Press the same shortcut again to stop. The final transcript is prefilled into Copilot Chat — review and send when ready.
5. Press Escape while recording to cancel and discard.

Optional: set copilotSpeech.language if you are not dictating in English. The first run downloads the local Moonshine model for that language.

How it works

The extension coordinates a local helper process instead of loading microphone or inference code into the extension host.

flowchart LR
	Command[Command or keybinding] --> Session[Dictation session]
	Session -->|bounded NDJSON| Helper[Native helper]
	Helper --> Capture[Microphone capture]
	Capture --> Moonshine[Moonshine streaming ASR]
	Moonshine -->|partial and final events| Helper
	Helper -->|versioned events| Session
	Session --> Delivery[Chat delivery adapter]
	Delivery -->|prefill only| Chat[Copilot Chat]

The helper owns raw PCM, capture, voice activity detection, and inference. This keeps audio outside the extension host, prevents a helper crash from taking down VS Code, and avoids Electron or Node native-addon ABI coupling.

Reference

License

MIT