Qwen3-TTS Technical Report
Paper · https://arxiv.org/abs/2601.15621
In this report, we present the Qwen3-TTS series, a family of advanced multilingual, controllable, robust, and streaming text-to-speech models. Qwen3-TTS supports state-of-the-art 3-second voice cloning and description-based control, allowing both the creation of entirely novel voices and fine-grained manipulation over the output speech. Trained on over 5 million hours of speech data spanning 10 languages, Qwen3-TTS adopts a dual-track LM architecture for real-time synthesis, coupled with two speech tokenizers: 1) Qwen-TTS-Tokenizer-25Hz is a single-codebook codec emphasizing semantic content, which offers seamlessly integration with Qwen-Audio and enables streaming waveform reconstruction via a block-wise DiT. 2) Qwen-TTS-Tokenizer-12Hz achieves extreme bitrate reduction and ultra-low-latency streaming, enabling immediate first-packet emission (97,ms) through its 12.5 Hz, 16-layer multi-codebook design and a lightweight causal ConvNet. Extensive experiments indicate state-of-the-art performance across diverse objective and subjective benchmark (e.g., TTS multilingual test set, InstructTTSEval, and our long speech test set). To facilitate community research and development, we release both tokenizers and models under the Apache 2.0 license.
First seen by Oriel on .
AI summary
The Qwen3-TTS technical report introduces multilingual, controllable, robust, streaming text-to-speech models with voice cloning, description-based control, and two specialized speech tokenizers.
- Qwen3-TTS is trained on over 5 million hours of speech spanning 10 languages and supports 3-second voice cloning.
- Its dual-track LM architecture targets real-time synthesis, including novel voice creation and fine-grained output speech manipulation.
- Qwen-TTS-Tokenizer-25Hz uses a single-codebook codec and block-wise DiT for streaming waveform reconstruction and integration with Qwen-Audio.
- Qwen-TTS-Tokenizer-12Hz uses a 12.5 Hz, 16-layer multi-codebook design and causal ConvNet to emit the first packet in 97 milliseconds.
- The models and both tokenizers are released under the Apache 2.0 license for community research and development.
#text-to-speech #voice cloning #speech tokenizers #streaming synthesis
Generated by gpt-5.6-sol. Check the original for details.
Metric trends
Last 30 days · Up to 120 recorded samples per source and metric, with reposts kept separate · First and latest retained
Points
Score reported by the source.
- Hugging Face Papers: 81, latest observation ; +1 over observed period since
Comments
Comment count reported by the source.
- Hugging Face Papers: 5, latest observation ; 0 over observed period since
Aliases
| Source | External ID | URL |
|---|---|---|
| Hugging Face Papers | 2601.15621 | https://arxiv.org/abs/2601.15621 Read discussion |