MiniMax TTS Skill

Generate short audio via MiniMax TTS (default: <20 seconds).

This skill is not restricted to Cantonese. Cantonese (Yue) is the default profile only. If the user requests another language (e.g. 國語 / Mandarin, 英語 / English), generate in that language and choose appropriate --voice / --boost values.

Quick use

Create an mp3 (default Cantonese profile):

{baseDir}/scripts/tts.sh --text "今晚食咩好？" --output "./out.mp3"

Recommended Cantonese defaults (already set in the script):

• --boost "Chinese,Yue"
• --voice "Chinese (Mandarin)_HK_Flight_Attendant" (HK-accented voice; adjust if you have a better voice_id)
• These are defaults, not a restriction. For 國語 / Mandarin, 英語 / English, or other languages, override --voice and (if useful) --boost.

Workflow (important)

• 先寫好文字，再生成語音 (draft text first, then generate audio).
• Always review the exact text in a readable form before TTS. The text should visibly include punctuation, spacing, pauses, ellipses, and tone/attitude cues.
• If the text does not read well on screen, the generated speech usually sounds worse.
• Practical writing cues for better output:
  • punctuation for pauses: ， 。 ？ ！
  • spacing / line breaks for rhythm and emphasis
  • ellipses for hesitation / soft pause: ... or ……
  • interjections / tone words when needed (e.g. 啊, 啦, 喎)

Configuration

Provide MINIMAX_API_KEY via OpenClaw skill env (recommended) or a local .env file.

Option B: .env next to the skill

Create:

• {baseDir}/.env

With:

MINIMAX_API_KEY=...

Notes / guardrails

• Keep requests short. For the “<20s” default, aim for roughly ⇐ 80–120 Chinese characters (depends on speed and punctuation).
• Use punctuation to control pauses.
• Follow the user’s requested language. Override the default Cantonese-oriented --voice / --boost when generating 國語 / Mandarin, 英語 / English, or other languages.
• Store generated files in an organized folder (not project root). Keep the source text and mp3 together using the same base name.
• Recommended pattern: generated/minimax-tts/YYYY-MM-DD/HHMMSS-topic-v01.txt and .mp3
• Create the output directory first (mkdir -p generated/minimax-tts/YYYY-MM-DD) before running tts.sh.
• If you need a different voice, pass --voice <voice_id>.