opencode-voice

Speech-to-text plugin for OpenCode — speak into your microphone and see your words appear in the prompt field in real-time.

Features

• 🎙️ Real-time transcription — words appear as you speak
• 🔌 3 providers — Deepgram (streaming), Groq Whisper (fast), OpenAI Whisper
• 🖥️ Cross-platform — macOS, Linux, Windows
• ⌨️ Simple toggle — Ctrl+Shift+V to start/stop
• 👁️ Live preview — interim text shown in overlay, final text inserted in prompt
• ✏️ Review before send — text goes to prompt for editing, never auto-sent

Requirements

A recording tool must be installed on your system:

| Platform | Recommended | Install | | -------- | ----------- | ---------------------------------------------------- | | macOS | SoX | brew install sox | | Linux | SoX | sudo apt install sox | | Windows | SoX | winget install sox or choco install sox.portable |

Fallback: FFmpeg is also supported on macOS and Linux.

Installation

Add to your OpenCode config (~/.config/opencode/opencode.json):

{
  "plugin": [["opencode-voice", { "provider": "deepgram" }]]
}

Or without options (configure via environment variables only):

{
  "plugin": ["opencode-voice"]
}

Configuration

Plugin Options (in opencode.json)

| Option | Type | Default | Description | | ----------------- | ---------------------------------------------- | ---------------------------- | ----------------------------------- | | provider | "deepgram" | "groq" | "openai-whisper" | — | STT provider to use | | language | string | auto-detect | Language code (e.g. "en", "fr") | | chunkDurationMs | number | 5000 (Groq) / 10000 (OpenAI) | Chunk size for HTTP providers |

Environment Variables

| Variable | Description | | ------------------------- | -------------------------------------------------------- | | OPENCODE_VOICE_PROVIDER | Override provider (takes precedence over plugin options) | | OPENCODE_VOICE_LANGUAGE | Override language | | DEEPGRAM_API_KEY | Deepgram API key | | GROQ_API_KEY | Groq API key | | OPENAI_API_KEY | OpenAI API key |

Security: API keys are read from environment variables only — never stored in config files.

Usage

1. Press Ctrl+Shift+V to start recording
2. Speak — you'll see a ● Recording... indicator with live preview
3. Press Ctrl+Shift+V again to stop
4. The transcribed text appears in the prompt field
5. Review/edit, then press Enter to send

Providers

| Provider | Protocol | Latency | Interim Results | Best For | | ------------------ | --------- | ------- | ------------------ | ------------------- | | Deepgram | WebSocket | ~100ms | ✅ Yes | Real-time streaming | | Groq | HTTP | ~200ms | ❌ No (5s chunks) | Speed + cost | | OpenAI Whisper | HTTP | ~500ms | ❌ No (10s chunks) | Accuracy |

Deepgram (Recommended)

Best real-time experience. Uses WebSocket streaming with interim results.

["opencode-voice", { "provider": "deepgram" }]

Get a free API key at deepgram.com.

Groq Whisper

Ultra-fast HTTP transcription (189x realtime). Good balance of speed and cost.

["opencode-voice", { "provider": "groq" }]

Get a free API key at console.groq.com.

OpenAI Whisper

Most widely used. Requires an OpenAI API key.

["opencode-voice", { "provider": "openai-whisper" }]

Troubleshooting

"No recording tool found"

Install SoX for your platform (see Requirements above).

"Invalid API key"

Check that your API key environment variable is set correctly:

echo $DEEPGRAM_API_KEY  # or GROQ_API_KEY / OPENAI_API_KEY

No microphone input

• Check that your microphone is connected and set as the default input device
• On Linux, ensure PulseAudio is running: pulseaudio --check
• On macOS, grant microphone permissions to your terminal app

Text appears in wrong position

The transcribed text is inserted at the cursor position in the prompt. If you've typed text before recording, the transcription will be appended after it.

Roadmap (v2)

• Google Cloud Speech (gRPC streaming)
• Local Whisper (whisper.cpp, no API key needed)
• OpenAI Realtime API (WebSocket, ultra-low latency)

License

MIT