opencode-audio

An OpenCode plugin that adds AI audio generation tools — music via Suno and text-to-speech via ElevenLabs and Fish Audio.

Tools

Tool	Description
generate_music	Generate AI music with Suno (async polling, saves .mp3)
download_music	Download previously generated Suno music by task ID
generate_voice_elevenlabs	Text-to-speech with ElevenLabs (multilingual)
list_elevenlabs_voices	Browse available ElevenLabs voices
generate_voice_fish	Text-to-speech with Fish Audio (best for Chinese)
list_fish_voices	Browse available Fish Audio voice models

Install

npm install opencode-audio

Add to your opencode.json:

{
  "plugins": {
    "audio": "opencode-audio"
  }
}

Local development

Clone this repo into your project and point to it:

{
  "plugins": {
    "audio": "./path/to/opencode-audio"
  }
}

Configuration

Set API keys as environment variables (e.g. in .env):

# Music generation (Suno)
SUNO_API_KEY=your-key-here        # https://sunoapi.org

# Voice generation (use one or both)
ELEVENLABS_API_KEY=your-key-here  # https://elevenlabs.io
FISH_API_KEY=your-key-here        # https://fish.audio

Each tool requires its corresponding API key to be set as an environment variable. Keys are read once when the plugin loads.

Usage

Once installed, the tools are available to the AI assistant automatically.

Generate music

Generate epic orchestral background music, save to assets/audio/

The generate_music tool accepts:

Parameter	Required	Description
track_id	Yes	Output filename (without extension)
prompt	Yes	Text description of the music
style	Yes	Genre/style tags (e.g. "epic orchestral, cinematic, 90 bpm")
output_dir	No	Output directory relative to project root
instrumental	No	Instrumental only, no vocals (default: true)
model	No	Suno model: "V4", "V4_5", "V4_5PLUS", "V4_5ALL", "V5" (default: "V4_5ALL")

Download music

Download the music from task abc123, save to assets/audio/

Use download_music to retrieve tracks from a previous generate_music call — especially useful if the original request timed out. Suno generates 2 tracks per request.

Parameter	Required	Description
task_id	Yes	Task ID returned from a previous generate_music call
filename	Yes	Output filename (without extension)
output_dir	No	Output directory relative to project root
track_index	No	Which track to download, 0 or 1 (default: 0)

Text-to-speech (ElevenLabs)

List available voices, then generate "Welcome aboard" with a deep male voice

Parameter	Required	Description
text	Yes	Text to convert to speech
voice_id	Yes	ElevenLabs voice ID (find via list_elevenlabs_voices)
filename	Yes	Output filename (without extension)
output_dir	No	Output directory relative to project root
model_id	No	TTS model (default: "eleven_multilingual_v2")
stability	No	Voice stability 0.0–1.0 (default: 0.5)
similarity_boost	No	Similarity boost 0.0–1.0 (default: 0.75)
style	No	Style exaggeration 0.0–1.0 (default: 0.4)
format	No	"mp3_44100_128", "opus", "pcm_44100" (default: "mp3_44100_128")

Text-to-speech (Fish Audio)

Generate Chinese speech "你好世界" using Fish Audio

Parameter	Required	Description
text	Yes	Text to convert to speech
reference_id	Yes	Fish Audio voice model ID (find via list_fish_voices)
filename	Yes	Output filename (without extension)
output_dir	No	Output directory relative to project root
format	No	"mp3", "wav", "opus" (default: "mp3")
bitrate	No	Audio bitrate in kbps (default: 128)

Development

npm install
npm run typecheck   # Type-check without emitting
npm run build       # Compile to dist/

License

MIT