Voice Typing — VS Code / Cursor Extension

Speak into your microphone, get a clean coding prompt inserted into your editor. Everything runs locally — no cloud services required.

Flow: Mic → whisper.cpp (local STT) → Ollama (local LLM rewrite) → Insert into editor

Platforms: macOS · Linux · Windows — works in both VS Code and Cursor

Install

From Marketplace: Voice Typing on VS Code Marketplace

Or from VSIX: Download the latest .vsix from GitHub Releases, then in VS Code/Cursor: Cmd+Shift+P → "Install from VSIX"

Quick Start (macOS)

brew install whisper-cpp sox       # STT engine + audio recorder
brew install ollama                # optional: LLM rewrite
ollama pull llama3.2:3b            # optional: pull rewrite model

Install the extension from the VS Code Marketplace, or build from source:

git clone https://github.com/bread22/voice-typing.git
cd voice-typing && npm install && npm run package
# Then: Cmd+Shift+P → "Install from VSIX" → select voice-prompt-*.vsix

Press Option+V to start recording. The whisper model (~75 MB) downloads automatically on first use.

Cross-Platform Installation

Step 1: Install whisper-cpp (speech-to-text engine)

brew install whisper-cpp

whisper-cli --help

sudo apt install build-essential libsdl2-dev
git clone https://github.com/ggerganov/whisper.cpp.git
cd whisper.cpp
cmake -B build
cmake --build build --config Release
sudo cp build/bin/whisper-cli /usr/local/bin/

whisper-cli --help

git clone https://github.com/ggerganov/whisper.cpp.git
cd whisper.cpp
cmake -B build
cmake --build build --config Release

whisper-cli.exe --help

Step 2: Install an audio recorder

The extension auto-detects the best available recorder for your platform:

Detection priority:

If no recorder is found, the extension shows the appropriate install command.

Step 3: Install Ollama (optional — LLM rewrite)

Ollama rewrites your raw voice transcript into a clean, structured prompt. Without it, the raw transcript is inserted directly (still useful!).

brew install ollama
ollama pull llama3.2:3b

curl -fsSL https://ollama.com/install.sh | sh
ollama pull llama3.2:3b

ollama pull llama3.2:3b

Step 4: Install the extension

git clone https://github.com/bread22/voice-typing.git
cd voice-typing
npm install
npm run package

This creates a voice-prompt-*.vsix file. Install it:

• VS Code: Ctrl+Shift+P → "Extensions: Install from VSIX..." → select the file
• Cursor: Cmd+Shift+P (Mac) / Ctrl+Shift+P (Win/Linux) → "Extensions: Install from VSIX..." → select the file

Reload the editor when prompted.

Step 5: First run

1. Press Alt+V (or Option+V on Mac)
2. On first use, the whisper model (ggml-tiny.en.bin, ~75 MB) downloads automatically
3. Grant microphone access if prompted by your OS
4. Speak naturally — the extension auto-stops when you pause

Usage

Start recording

Recording flow

1. Press shortcut → status bar shows "Recording..."
2. Speak naturally — e.g. "add a function that validates email addresses using regex"
3. Pause speaking → silence detection (VAD) auto-stops recording
4. Status bar progress: Listening → Transcribing → Rewriting → Done
5. Result inserted at your cursor position (cursor moves to end, space appended)

Example

You say:

"um so I need a function that uh validates email addresses and it should use regex and return true or false"

Gets inserted as:

Add a function that validates email addresses using regex. It should return true for valid emails and false otherwise.

Without Ollama (no rewrite)

If Ollama isn't running, the raw transcript is inserted directly. You can also disable rewrite explicitly by setting voicePrompt.rewrite.provider to none.

Settings Reference

Open settings (Ctrl+,) and search for voicePrompt.

Speech-to-Text

Model sizes (auto-downloaded):

Rewrite (LLM)

Behavior

Voice Activity Detection (VAD)

VAD tuning tips:

• Getting cut off mid-sentence? Increase silenceMs by 100–200 ms
• Feels slow to respond? Decrease silenceMs by 100 ms
• Ambient noise triggering? Increase minSpeechMs to 400–500 ms

Cloud Rewrite (Optional)

Use OpenAI or another cloud LLM instead of (or as fallback to) local Ollama:

1. Set API key: Ctrl+Shift+P → Voice Prompt: Set Cloud API Key (stored securely in VS Code SecretStorage)
2. Switch provider: Set voicePrompt.rewrite.provider to cloud
3. Or use as fallback: Keep ollama as provider, enable voicePrompt.autoFallbackToCloud

Troubleshooting

Architecture

┌─────────────┐    ┌──────────┐    ┌─────────────┐    ┌───────────┐    ┌─────────┐
│  Microphone  │ →  │ Platform │ →  │ whisper-cpp │ →  │  Ollama   │ →  │ Editor  │
│  (Alt+V)     │    │ Recorder │    │ (local STT) │    │ (rewrite) │    │ Insert  │
└─────────────┘    │ WAV file │    │ Raw text    │    │ Clean text│    │ @ cursor│
                   └──────────┘    └─────────────┘    └───────────┘    └─────────┘

All processing happens locally on your machine. No audio or text is sent to any cloud service unless you explicitly configure a cloud rewrite provider.

Project Structure

src/
  extension.ts           — command registration and activation
  audio/
    audioCaptureService  — cross-platform mic capture (SoX / arecord / FFmpeg) + VAD
  stt/
    whisperCppSttProvider — shells out to whisper-cli binary
    httpSttProvider        — HTTP STT for custom servers (opt-in)
    modelManager           — auto-downloads GGML models from HuggingFace
  rewrite/
    ollamaRewriteProvider  — local Ollama LLM rewrite
    cloudRewriteProvider   — cloud LLM rewrite (OpenAI-compatible)
  inject/
    cursorInputInjector    — inserts text at cursor, manages position
  config/
    settings               — reads VS Code configuration
    secrets                — manages API keys via SecretStorage
  types/
    contracts              — shared TypeScript interfaces
  orchestration/
    voicePromptOrchestrator — pipeline: record → transcribe → rewrite → inject

Development

npm install          # install dependencies
npm run build        # compile TypeScript
npm run bundle:dev   # esbuild development bundle (with source maps)
npm run bundle       # esbuild production bundle (minified)
npm run package      # create .vsix for distribution
npm run dev          # build + package dev vsix
npm run watch        # watch mode for development

Press F5 in VS Code/Cursor to launch the Extension Development Host with the debugger attached.

License

MIT