happy-audio-gen

happy-audio-gen

Turns text into speech across 6 providers through one CLI. All providers are synchronous (TTS is fast — typically under 10 seconds) except Bailian's voice-design flow (which is still covered but uses a longer poll window).

Quick usage

# Shortest path — OpenAI default voice
bun scripts/main.ts --text "Hello, world" --out ./hello.mp3

# Chinese, MiniMax
bun scripts/main.ts --provider minimax --text "大家好" --voice male-qn-qingse --out ./hello.mp3

# Long-form, Bailian (auto-splits by sentence)
bun scripts/main.ts --provider bailian --textfiles ./script.md --out ./narration.mp3

When to invoke this skill

• User asks to synthesize speech / TTS / read aloud / narrate / dub / make a voice-over.
• User asks to convert script / text / article into audio.
• User names a TTS voice or model.

Do not route here when the user wants to transcribe audio → text (that's STT, different domain), or edit / mix audio files (use a dedicated audio editor).

Step 0: Preflight (BLOCKING)

1. Locate EXTEND.md:

   • ./.happy-skills/happy-audio-gen/EXTEND.md
   • $XDG_CONFIG_HOME/happy-skills/happy-audio-gen/EXTEND.md
   • ~/.happy-skills/happy-audio-gen/EXTEND.md

   If none found, run bun scripts/main.ts --setup and walk the user through references/config/first-time-setup.md.

2. Verify at least one provider has credentials (env var or 1Password reference).

3. Verify Bun is available. Fallback: npx -y bun.

Step 1: Choose provider

Preference order:

1. --provider <id>
2. EXTEND.md default_provider
3. Auto-detect env vars: openai > elevenlabs > bailian > minimax > siliconflow > playht

Pick by language / voice intent:

• English, natural + fast → openai (gpt-4o-mini-tts / tts-1).
• Multilingual, voice cloning → elevenlabs.
• Chinese, long-form → bailian (qwen-tts auto-chunks long scripts) or minimax.
• Chinese dialect / voice design → bailian (voice-design with qwen3-tts-vd) or siliconflow (CosyVoice2).
• Ultra-realistic, short-form → playht (2.0).