Voice Clone + TTS

Scope

This skill is narrowly scoped to: (1) uploading clone audio to MiniMax, (2) creating a cloned voice, (3) TTS with cloned or existing voices, and (4) updating the cloned-voice mapping block in this SKILL.md. The script only reads/writes this skill’s SKILL.md; it does not read unrelated system files or other environment variables beyond the MiniMax API key(s) above.

When to Use

• Use this skill when you need to clone user-provided audio into a reusable voice.
• Use this skill when you need text-to-speech (TTS) with an already cloned voice.
• For long-term maintenance, cloned results are written back to this file to reduce repeated setup.

API Reference (MiniMax)

• Upload clone audio: POST /v1/files/upload, purpose=voice_clone, multipart/form-data.
• Create cloned voice: POST /v1/voice_clone.
• Speech synthesis: POST /v1/t2a_v2.
• Audio requirements: mp3/m4a/wav, duration 10 seconds–5 minutes, file size <=20MB.

Install

• Runtime: Python 3.7+.
• Dependencies: The script uses the requests library. Install with:

  pip install -r requirements.txt
  

  or pip install requests.
• Network: The script calls MiniMax APIs over HTTPS; it does not read unrelated system files. It only reads/writes this skill’s SKILL.md to update the cloned-voice mapping block.

Required environment variables (credentials)

At least one of the following must be set for MiniMax API authentication (see frontmatter requiredEnv / optionalEnv):

The script will fail with a clear error if none are set.

Prerequisites

1. Credentials: Set one of the env vars above (see “Required environment variables”).
2. Prepare clone audio (format/duration/size limits above).
3. Before cloning, confirm the voice name (voice_name) with the user, e.g. liuyang_narration_v1.

Usage

1. Go to the skill directory: cd workspace/skills/voice-clone-tts
2. Run the script (clone + synthesize):

python scripts/minimax_voice_clone_tts.py \
  --audio "/absolute/path/to/voice.wav" \
  --voice-name "yangtuo_demo_v1" \
  --display-name "Alpaca Demo" \
  --text "Hello, this is a cloned voice test." \
  --output "./output/voice_test.mp3"

3. To clone only (no synthesis), omit --text.
4. To synthesize only (by display name or voice_id):

# Resolve by display name
python scripts/minimax_voice_clone_tts.py \
  --voice "voice_v2" \
  --text "This is TTS using an existing cloned voice." \
  --output "./output/reuse_voice.mp3"

# Or specify voice_id directly
python scripts/minimax_voice_clone_tts.py \
  --voice-id "yangtuo_demo_v1" \
  --text "This is TTS using an existing cloned voice." \
  --output "./output/reuse_voice.mp3"

Common Options

• --audio: Path to clone audio (required for cloning).
• --voice-name: Required when cloning; API voice ID (letters, digits, underscores, e.g. yangtuo_demo_v1).
• --display-name: Optional when cloning; display name written to SKILL (e.g. Alpaca Demo). Defaults to --voice-name if omitted.
• --voice-id: For synthesis, specify API voice_id directly (skips mapping table).
• --voice: For synthesis, specify display name or voice_id; resolved from the mapping table below (e.g. voice_v2 or yangtuo_demo_v1).
• --text: Text to synthesize (omit for clone-only).
• --output: Output audio path (default ./output/minimax_tts.mp3).
• --model: Speech model (default speech-2.8-turbo).
• --format: Output format (mp3/pcm/flac/wav).
• --speed --vol --pitch --emotion: Speech expression parameters.

Write-Back (Important)

• After a successful clone, the script writes display name → voice_id to the “Cloned Voice Mapping” section below.
• Use a display name with --voice "display name" so you don't need to remember voice_id.

Cloned Voice Mapping

• Left: display name; right: API voice_id. For TTS use --voice "display name" or --voice-id voice_id.

• test_voice_1772187110: test_voice_1772187110 (updated: 2026-02-27 18:12:00)
• voice_v1: shuangyue_test (updated: 2026-02-28 16:47:01)
• voice_v2: yangtuo_demo_v1 (updated: 2026-02-27 18:19:39)
• voice_v3: dong_yuhui_voice_v1 (updated: 2026-03-02 19:51:44)

Troubleshooting

• 401 / auth failure: Check that MINIMAX_API_KEY is correct.
• Parameter errors: Check voice_name rules, audio format/size, and text length.
• Clone succeeded but not written back: Ensure SKILL.md exists and contains the write-back marker block.