Qwen3-TTS

Voice Design, Clone, and Generation

Qwen3-TTS introduction

Qwen3-TTS is the leading open-source text-to-speech model featuring zero-shot voice cloning, emotional control, and support for 10+ languages. Generate natural, human-like speech with ultra-low latency.

Website:: qwen3-tts.app

📝 🔉 Text To Speech

Qwen3-TTS overview

Qwen3-TTS is an open-source text-to-speech model for natural voice synthesis. It uses a high-efficiency 12Hz tokenizer and multi-codebook speech encoder to balance sample compression and detail retention, capturing paralinguistic features like breath, hesitation, and emotional intensity. It supports zero-shot voice cloning from a 3-second reference clip, context-aware prosody, multilingual synthesis in over 10 languages, and real-time streaming with 97ms first token latency. It offers Python SDK and OpenAI-compatible API for integration.

Qwen3-TTS features

Zero-shot voice cloning from a 3-second reference clip
Emotional control and natural language audio control (whisper, shout, laugh, speak fast)
Multilingual synthesis in over 10 languages including English, Chinese, Japanese, Korean, French, and German
Context-aware prosody and intonation
High-efficiency 12Hz tokenizer
Multi-codebook speech encoder
Long-form audio synthesis
Real-time streaming with 97ms first token latency
Python SDK and OpenAI-compatible API
Docker deployment for OpenAI-compatible API server
Apache 2.0 open-source license
Code-switching support