Sonic

Generate voice, music, lyrics, and sound effects through scripts/sonic.js — a zero-dependency Node.js CLI (Node 18+, no npm install). It saves audio to disk and prints the absolute path.

Quick start

# TTS (defaults to SiliconFlow CosyVoice2)
node scripts/sonic.js tts "Hello, world" -o hi.mp3

# Music with auto-written lyrics / instrumental only
node scripts/sonic.js music "cheerful jazz with brass and double bass, café style" -o song.mp3 --lyrics-optimizer
node scripts/sonic.js music "epic movie soundtrack, orchestral" -o bgm.mp3 --instrumental

# Lyrics
node scripts/sonic.js lyrics "a summer day at the beach"

# Sound effects — cloud MMAudio (zero install, needs key) / local Woosh (free, auto-installed)
node scripts/sonic.js sfx "rain falling on a wooden roof" -o rain.wav --provider mmaudio
node scripts/sonic.js sfx "footsteps crunching in the snow" -o steps.flac

Model selection

China-friendly priority order (Alipay/WeChat top-up, no foreign card):

Asset	Recommended (priority order)	Notes
Voice / TTS	SiliconFlow CosyVoice2 → MiniMax Speech → OpenAI TTS / ElevenLabs	SiliconFlow is OpenAI-compatible and already keyed
Music	MiniMax Music 3.0 → Bailian Fun-Music → Suno (third-party relays)	MiniMax has an official API + China payment
Sound effects	Local MMAudio → Local Sony Woosh → MMAudio cloud → ElevenLabs SFX	Local channels run on an NVIDIA GPU, free, auto-installed by the agent

Key workflows

Local SFX prompts must be English — MMAudio and Woosh use English CLIP/CLAP text encoders; non-English prompts produce garbled, voice-like artifacts. Add --negative "speech, voice, human talking, vocals, singing" to suppress them.
Music prompts: {genre/style} + {instruments} + {mood} + {tempo} + {scene}; --instrumental for pure music, --lyrics-optimizer or --lyrics for vocals. Generation takes tens of seconds to minutes — use a generous timeout.
TTS: --speed 0.9–1.1 for narration, 1.2+ for ads, 0.8 for audiobook; -v <voice> keeps a consistent voice. ElevenLabs' free tier only ships its 21 premade voices (default "Adam") — list them with GET /v1/voices.
Local channels auto-install: when local Woosh / MMAudio is needed but missing, run scripts/install/woosh.* / scripts/install/mmaudio.* with -Start / --start yourself — never ask the user to install by hand. Servers idle-exit after 15 minutes; restart before retrying a failed request.
Woosh's API server serves Woosh-DFlow (fast) only; quality and video-to-audio modes need the Gradio UI.

References — load on demand

When	Reference
Full CLI options and exit codes	core-usage
First run with no API key configured (env keys, AGENTS.md merge)	core-setup
Debugging the pipeline or editing the scripts	core-script