clone-voicelisted

# Clone Voice / 克隆声音 ## When to use / 何时使用 Project setup, when the user supplies a reference voice sample (1–3 minutes clean speech). 项目初始化阶段，用户提供 1-3 分钟干净语音样本时使用。 ## Inputs / 输入 - `inputs/voice_sample.wav` or `.mp3` (mono, 44.1k+, <10 MB). - `--name` voice label. - `ELEVENLABS_API_KEY` env. ## Output / 输出 - `config/voice.json`: `{voice_id, name, created_at, sample_path, similarity_baseline}`. ## Steps / 步骤 1. Validate sample: length 60–180s, mono, no clipping (peak < -1 dBFS). 2. Call `mcp__elevenlabs__voices_add` with name + sample bytes. 3. Persist returned `voice_id` to `config/voice.json`. 4. Run a 10-word sanity TTS via `test-voice-sample`. ## Examples / 示例 ``` claude run clone-voice --name "Hiro-EN" --sample inputs/voice_sample.wav ``` ## Failure recovery / 失败回退 - If sample fails validation, run ffmpeg loudnorm + downmix and retry once. - If quota exhausted, fall back to a stock voice and log to `workspace/logs/voice.fallback.log`.