Voice Clone Skill

Use this skill to clone a speaker's voice and generate text-to-speech audio.

Two-Step Process

Step 1: Clone Voice (one-time)

python skills/voice-clone/clone.py <audio_sample.wav> [--transcript "text"]

Creates a speaker embedding file that can be reused.

Step 2: Generate Speech

python skills/voice-clone/speak.py <embedding.safetensors> "Text to speak"

Generates audio using the cloned voice.

Requirements

• FAL_KEY in .env (fal.ai API key)
• Voice sample: 10-30 seconds of clear speech (WAV/MP3)
• Optional: Transcript of the sample for better quality

Output

• assets/outputs/voice_embeddings/<name>_embedding.safetensors - Reusable voice model
• assets/outputs/audio/<name>_speech.wav - Generated audio

Notes

• qwen3-tts works best with Chinese speech samples
• Cross-lingual cloning (Chinese voice → English speech) may have quality variations
• Provide reference transcript for best quality