Community

Blog Events Webinars Tutorials Forum
  1. Blog
  2. Events
  3. Webinars
  4. Tutorials
  5. Forum
Create Account
×
Community Blog Qwen3-TTS Family is Now Open Sourced: Voice Design, Clone, and Generation!

Qwen3-TTS Family is Now Open Sourced: Voice Design, Clone, and Generation!

This article introduces the open-sourced Qwen3-TTS family, a high-performance speech generation model series supporting voice cloning, design, multilingual synthesis, and ultra-low-latency streaming.

Qwen3-TTS is a series of powerful speech generation capabilities developed by Qwen, offering comprehensive support for voice clone, voice design, ultra-high-quality human-like speech generation, and natural language-based voice control. It provides developers and users with the most extensive set of speech generation features available. Powered by the innovative Qwen3-TTS-Tokenizer-12Hz multi-codebook speech encoder, Qwen3-TTS achieves efficient compression and robust representation of speech signals. This not only fully preserves paralinguistic information and acoustic environmental features but also enables high-speed, high-fidelity speech reconstruction via a lightweight non-DiT architecture. Utilizing Dual-Track modeling, Qwen3-TTS achieves extreme bidirectional streaming generation speeds, where the first audio packet is delivered after processing just a single character. The entire Qwen3-TTS multi-codebook model series is now open-sourced, featuring two sizes: 1.7B and 0.6B. The 1.7B model delivers peak performance and powerful control capabilities, while the 0.6B model offers an ideal balance between performance and efficiency. The models support 10 mainstream languages (Chinese, English, Japanese, Korean, German, French, Russian, Portuguese, Spanish, and Italian) along with various dialects to meet global application demands. Furthermore, the models exhibit strong contextual understanding, allowing them to adapt tone, rhythm, and emotional expression based on instructions and text semantics, while significantly improving robustness to input text noise. Now open-sourced on GitHub and accessible via the Qwen API.

Model List

1.7B Model

Model Features Language Support Streaming Instruction Control
Qwen3-TTS-12Hz-1.7B-VoiceDesign Performs voice design based on user-provided descriptions. Chinese, English, Japanese, Korean, German, French, Russian, Portuguese, Spanish, Italian ✅ ✅
Qwen3-TTS-12Hz-1.7B-CustomVoice Provides style control over target timbres via user instructions; supports 9 premium timbres covering various combinations of gender, age, language, and dialect. Chinese, English, Japanese, Korean, German, French, Russian, Portuguese, Spanish, Italian ✅ ✅
Qwen3-TTS-12Hz-1.7B-Base Base model capable of 3-second rapid voice clone from user audio input; can be used for fine-tuning (FT) other models. Chinese, English, Japanese, Korean, German, French, Russian, Portuguese, Spanish, Italian ✅

0.6B Models

Model Features Language Support Streaming Instruction Control
Qwen3-TTS-12Hz-0.6B-CustomVoice Supports 9 premium timbres covering various combinations of gender, age, language, and dialect. Chinese, English, Japanese, Korean, German, French, Russian, Portuguese, Spanish, Italian ✅
Qwen3-TTS-12Hz-0.6B-Base Base model capable of 3-second rapid voice clone from user audio input; can be used for fine-tuning (FT) other models. Chinese, English, Japanese, Korean, German, French, Russian, Portuguese, Spanish, Italian ✅

Qwen3-TTS Key Features

Main Features:

  • Powerful Speech Representation: Powered by the self-developed Qwen3-TTS-Tokenizer-12Hz, it achieves efficient acoustic compression and high-dimensional semantic modeling of speech signals. It fully preserves paralinguistic information and acoustic environmental features, enabling high-speed, high-fidelity speech reconstruction through a lightweight non-DiT architecture.
  • Universal End-to-End Architecture: Utilizing a discrete multi-codebook LM architecture, it realizes full-information end-to-end speech modeling. This completely bypasses the information bottlenecks and cascading errors inherent in traditional LM+DiT schemes, significantly enhancing the model’s versatility, generation efficiency, and performance ceiling.
  • Extreme Low-Latency Streaming Generation: Based on the innovative Dual-Track hybrid streaming generation architecture, a single model supports both streaming and non-streaming generation. It can output the first audio packet immediately after a single character is input, with end-to-end synthesis latency as low as 97ms, meeting the rigorous demands of real-time interactive scenarios.
  • Intelligent Text Understanding and Voice Control: Supports speech generation driven by natural language instructions, allowing for flexible control over multi-dimensional acoustic attributes such as timbre, emotion, and prosody. By deeply integrating text semantic understanding, the model adaptively adjusts tone, rhythm, and emotional expression, achieving lifelike “what you imagine is what you hear” output.

1

Model Performance

We have conducted a comprehensive evaluation of Qwen3-TTS across dimensions such as voice clone, voice design, and control. The results demonstrate that it has achieved SOTA performance across multiple metrics. Specifically:

  • In voice design tasks: Qwen3-TTS-VoiceDesign outperformed the MiniMax-Voice-Design closed-source model in both instruction-following capability and generative expressiveness on the InstructTTS-Eval benchmark, while significantly leading other open-source models.
  • In voice control tasks: Qwen3-TTS-Instruct demonstrates single-speaker multilingual generalization with an average Word Error Rate (WER) of 2.34%. It also features the ability to maintain timbre while providing precise style control, achieving a score of 75.4% on InstructTTS-Eval. Furthermore, it shows exceptional long-form speech generation capabilities, with a WER of 2.36% (Chinese) and 2.81% (English) during continuous 10-minute synthesis.
  • In voice clone tasks: Qwen3-TTS-VoiceClone surpassed MiniMax and SeedTTS in speech stability for both Chinese and English cloning on Seed-tts-eval. On the TTS multilingual test set across 10 languages, it achieved an average WER of 1.835% and a speaker similarity of 0.789, outperforming MiniMax and ElevenLabs. Its cross-lingual voice clone capabilities also reached SOTA, surpassing CosyVoice3.

2
3

Tokenizer Performance

We evaluated Qwen-TTS-Tokenizer for speech reconstruction. Results on the LibriSpeech test-clean set demonstrate that it achieves SOTA performance across all key metrics. Specifically, in Perceptual Evaluation of Speech Quality (PESQ), Qwen-TTS-Tokenizer achieved scores of 3.21 and 3.68 in wideband and narrowband respectively, significantly leading similar tokenizers. In Short-Time Objective Intelligibility (STOI) and UTMOS, Qwen-TTS-Tokenizer achieved scores of 0.96 and 4.16, demonstrating superior reconstruction quality. In speaker similarity, Qwen-TTS-Tokenizer achieved a score of 0.95, significantly surpassing comparison models, indicating its near-lossless speaker information preservation capability.

4

Source

AI Artificial Intelligence Developers Open Source TTS AI model Generative AI GenAI Qwen Speech-to-Speech Model Qwen3-TTS Speech Generation
0 1 0
Share on

Read previous post:

Momentum: How Alibaba Cloud Is Leading the New AI Paradigm

Read next post:

Alibaba Cloud Earns Industry Leading Recognition

Alibaba Cloud Community

1,362 posts | 485 followers

You may also like

Comments

Alibaba Cloud Community

1,362 posts | 485 followers

Related Products

More Posts by Alibaba Cloud Community

See All