---
name: qwen-voice
description: Generate speech from text using local Qwen3-TTS. Use when the user asks to generate voice/speech, read text aloud, or do TTS.
---

## Usage

Generate speech locally using Qwen3-TTS on Apple Silicon. No API keys needed.

### Generate speech

Run: `/Users/yzlabmac/agentspace/voiceai/qwen_voice_skill/skill/run.sh generate --text "<text>" [--voice <name>] [--mode <mode>] [--emotion "<instruction>"] [--speed <float>]`

Parameters:
- `--text` (required): The text to speak
- `--mode`: "custom" (default), "design", or "clone"
- `--voice`: Speaker name. Auto-detected by language if omitted. Options: ryan, aiden, serena, vivian (English); vivian, serena, uncle_fu, dylan, eric (Chinese); ono_anna (Japanese); sohee (Korean)
- `--emotion`: Emotion/tone instruction, e.g. "Happy and excited", "Calm and professional". Only for custom/design modes.
- `--speed`: Speed multiplier (0.8=slow, 1.0=normal, 1.3=fast). Only for custom mode.

For clone mode, also provide:
- `--ref-audio`: Path to reference audio file
- `--ref-text`: Transcript of the reference audio

### List available voices

Run: `/Users/yzlabmac/agentspace/voiceai/qwen_voice_skill/skill/run.sh list-voices`

### Play audio file

Run: `/Users/yzlabmac/agentspace/voiceai/qwen_voice_skill/skill/run.sh play <file_path>`

After generating, always tell the user the output file path and offer to play it.