A unified Text-to-Speech demo featuring three powerful modes: Voice, Clone and Design
Frequently asked questions
What languages does Qwen3-TTS support?
Qwen3-TTS supports 10 languages: Chinese (including regional dialects), English, Japanese, Korean, German, French, Russian, Portuguese, Spanish, and Italian. It can also perform cross-language voice cloning, for example cloning an English voice to speak Korean fluently.
How does Qwen3-TTS voice cloning work?
Qwen3-TTS can accurately clone any target voice using as little as 3 seconds of reference audio. The model analyzes the vocal characteristics from the short sample and reproduces them in generated speech, even across different languages, making it highly practical for real-world applications.
What is voice design in Qwen3-TTS?
Voice design lets you create a completely new custom voice using plain-language text descriptions, such as 'a calm female narrator with a slight Southern drawl.' No reference audio is needed. The model interprets the description and generates a unique voice matching the specified tone, style, and personality.
What is the latency of Qwen3-TTS?
Qwen3-TTS achieves a first-audio-packet latency of just 97 milliseconds, enabled by its custom Qwen3-TTS-Tokenizer-12Hz tokenizer. This low-latency architecture makes it suitable for near-real-time applications such as live voice assistants, interactive storytelling, and streaming audio generation.
What are the available model sizes for Qwen3-TTS?
Qwen3-TTS is available in two sizes: a 1.7B parameter version optimized for maximum output quality and a 600M parameter version optimized for faster inference speed. Users can choose based on their priority between audio fidelity and processing efficiency.
Can Qwen3-TTS control emotion and speaking style?
Yes. Qwen3-TTS supports emotion and style control through natural language instructions embedded directly in the prompt. You can specify tone, emotion, speaking speed, and emphasis, for example requesting an excited delivery or a slow, calm narration, without any additional configuration files.
How many times has Qwen3-TTS been run on CRAISEE?
Qwen3-TTS has accumulated over 2.3 million runs on the CRAISEE platform, making it one of the most widely used TTS models available. This high usage reflects its versatility across use cases including content creation, app development, dubbing, and interactive media production.
Who created Qwen3-TTS and where can I use it?
Qwen3-TTS was developed by the Qwen team at Alibaba Cloud and is hosted on Replicate. It is accessible through CRAISEE, which provides a unified demo interface featuring three modes: Voice generation, Voice Cloning, and Voice Design, suitable for creators, developers, and storytellers.
