Qwen3-TTS

⚡ NEW: Production-Optimized Inference
This implementation includes 5 advanced GPU optimizations (Flash Attention 2, torch.compile, TF32, cuDNN benchmark, BFloat16) for up to 40% faster inference compared to baseline. Expected RTF: 0.65-0.70 on RTX 3090 (54% faster than real-time). See OPTIMIZATION_GUIDE.md for details.

This repository provides an OpenAI-compatible FastAPI server for Qwen3-TTS, enabling drop-in replacement for OpenAI's TTS API endpoints. Built on top of the powerful Qwen3-TTS model series developed by the Qwen team at Alibaba Cloud, it offers comprehensive support for voice clone, voice design, ultra-high-quality human-like speech generation, and natural language-based voice control.

✨ Features

🎯 OpenAI API Compatible - Drop-in replacement for POST /v1/audio/speech
⚡ Multiple Backends - Choose between official or vLLM-Omni backend for optimal performance
🌐 Multi-language Support - 10+ languages (Chinese, English, Japanese, Korean, German, French, Russian, Portuguese, Spanish, Italian)
🎨 Multiple Voice Options - 9 premium voices with various gender, age, and dialect combinations
📊 Multiple Audio Formats - MP3, Opus, AAC, FLAC, WAV, PCM
⚡ GPU-Accelerated - Optimized for NVIDIA GPU/CUDA deployment
🔧 Text Sanitization - Advanced text preprocessing for URLs, emails, special characters
🐳 Docker Ready - Streamlined Dockerfile for NVIDIA GPU deployment
🖥️ Web Interface - Dark-themed interactive demo UI

Backend Options

This implementation supports two backend engines:

Backend	Speed	Setup	Best For	Status
Official (default)	⚡⚡ Excellent	✅ Simple	All use cases, production-ready	✅ Stable
vLLM-Omni	⚡⚡⚡ Fast	⚠️ Python 3.12 + CUDA	High-throughput, low-latency	✅ Available

Official Backend: Uses the official Qwen3-TTS Python implementation. Recommended for most users.
vLLM-Omni Backend: Uses vLLM-Omni for optimized inference. Requires Python 3.12 and a dedicated Docker image. See VLLM_BACKEND_STATUS.md for details.

🚀 Performance Benchmarks

Performance comparison with Flash Attention 2 optimization on NVIDIA RTX 3090 (24GB VRAM). RTF = Real-Time Factor (lower is better, <1.0 means faster than real-time).

Flash Attention 2 Impact

Backend	RTF (Avg)	Latency	Flash Attn Impact	Recommendation
Official + Flash Attn 2 ⚡	0.87	7.28s	✅ +10% faster	🏆 Best Overall
Official (baseline)	0.97	8.49s	-	Good
vLLM-Omni	0.83	7.85s	-	Fast (no flash)
vLLM-Omni + Flash Attn 2	0.90	8.14s	⚠️ -8% slower	Not recommended

Key Findings:

✅ Flash Attention 2 improves Official backend by 10%
⚠️ Flash Attention 2 degrades vLLM-Omni by 8% (optimization conflict)
🏆 Official + Flash Attn 2 is the fastest and most reliable configuration

Detailed Results (with Flash Attention 2)

Test Case	Words	Official+Flash2	vLLM+Flash2	Official RTF	vLLM RTF
Short	2	1.15s	1.29s	0.95	0.97
Sentence	7	2.65s	3.39s	0.88	0.89
Medium	20	7.60s	7.59s	0.84	0.87
Long	36	17.71s	20.29s	0.81	0.87

Model: Qwen3-TTS-12Hz-1.7B-CustomVoice
GPU: NVIDIA GeForce RTX 3090 (24GB VRAM)
Test Method: 1 cold run + 5 warm runs per prompt
Docker Images: Built with Flash Attention 2

Production Recommendation: Use Official backend with Flash Attention 2 for best performance (RTF 0.87, ~15% faster than real-time).

See BENCHMARK_RESULTS.md for full details.

⚡ Performance Optimizations

The official backend includes several production-ready optimizations for maximum inference speed:

Optimization	Impact	Hardware Requirement	Status
Flash Attention 2	+10% faster	Ampere+ GPU (RTX 30xx/40xx)	✅ Enabled
torch.compile()	+20-30% faster	Any CUDA GPU	✅ Enabled
TF32 Precision	+3-5x matmul speed	Ampere+ GPU	✅ Enabled
cuDNN Benchmark	+5-10% faster	Any CUDA GPU	✅ Enabled
BFloat16	-50% VRAM	Turing+ GPU (RTX 20xx+)	✅ Enabled

Combined Effect: ~25-35% speedup over baseline (expected RTF: 0.65-0.70)

⚠️ Note: torch.compile() and cuDNN benchmarking require warmup. First 2-3 requests may be slower (~10-30s) while optimizations initialize.

📖 See OPTIMIZATION_GUIDE.md for detailed information about each optimization, how to enable/disable them, and troubleshooting tips.

🚀 Quick Start (API Server)

Using OpenAI Python Client

from openai import OpenAI

# Point to your local Qwen3-TTS server
client = OpenAI(
    base_url="http://localhost:8880/v1",
    api_key="not-needed"  # API key not required for local server
)

# Generate speech
response = client.audio.speech.create(
    model="qwen3-tts",
    voice="Vivian",  # Or: Ryan, Serena, Dylan, Eric, Aiden, etc.
    input="Hello! This is Qwen3-TTS speaking with an OpenAI-compatible API.",
    response_format="mp3",  # Options: mp3, opus, aac, flac, wav, pcm
    speed=1.0  # 0.25 to 4.0
)

response.stream_to_file("output.mp3")

Language-Specific Models

You can force a specific language by using language-suffixed model names. This overrides any language parameter passed from the client (useful for integration with open-webui):

# Force Spanish output regardless of language parameter
response = client.audio.speech.create(
    model="tts-1-es",  # Spanish
    voice="Vivian",
    input="Hola! Este es Qwen3-TTS."
)

# Force French output
response = client.audio.speech.create(
    model="tts-1-hd-fr",  # French (HD quality)
    voice="Vivian",
    input="Bonjour! Ceci est Qwen3-TTS."
)

Supported Language Codes:

tts-1-en or tts-1-hd-en - English
tts-1-zh or tts-1-hd-zh - Chinese
tts-1-ja or tts-1-hd-ja - Japanese
tts-1-ko or tts-1-hd-ko - Korean
tts-1-de or tts-1-hd-de - German
tts-1-fr or tts-1-hd-fr - French
tts-1-es or tts-1-hd-es - Spanish
tts-1-ru or tts-1-hd-ru - Russian
tts-1-pt or tts-1-hd-pt - Portuguese
tts-1-it or tts-1-hd-it - Italian

When using these language-specific models, any language parameter in the request will be ignored, ensuring consistent output in the specified language.

Web Interface

After starting the server, visit http://localhost:8880 for an interactive web demo:

📦 Deployment

Option 1: Using Conda (Recommended for Development)

# Create a fresh conda environment
conda create -n qwen3-tts python=3.12 -y
conda activate qwen3-tts

# Install the package with API dependencies
pip install -e ".[api]"

# Optional: Install FlashAttention 2 for better performance
pip install -U flash-attn --no-build-isolation

# Start the API server
python -m api.main
# Or use the convenience script
./start_server.sh

The server will start on http://0.0.0.0:8880 by default.

Environment Variables:

HOST - Server host (default: 0.0.0.0)
PORT - Server port (default: 8880)
WORKERS - Number of workers (default: 1)
CORS_ORIGINS - CORS origins (default: *)
TTS_BACKEND - Backend engine: official or vllm_omni (default: official)
TTS_MODEL_NAME - Override default model (optional)
TTS_WARMUP_ON_START - Warmup on startup: true or false (default: false)

Backend Selection:

# Use official backend (default)
export TTS_BACKEND=official
python -m api.main

# Use vLLM-Omni backend for faster inference
export TTS_BACKEND=vllm_omni
export TTS_WARMUP_ON_START=true
pip install -e ".[vllm]"  # Install vLLM first
python -m api.main

For detailed vLLM-Omni setup and configuration, see docs/vllm-backend.md.

Option 2: Using Docker (NVIDIA GPU)

# Build and run with GPU support
docker build -t qwen3-tts-api .
docker run --gpus all -p 8880:8880 qwen3-tts-api

# Or use Docker Compose for easier management
docker-compose up

Note: The Dockerfile is configured for NVIDIA GPU deployment. For vLLM-Omni backend (optional, faster), see docs/vllm-backend.md and use Dockerfile.vllm.

Docker Compose Configuration

The docker-compose.yml is configured for NVIDIA GPU with the official backend:

# Start the service
docker-compose up

# Run in background
docker-compose up -d

# View logs
docker-compose logs -f

# Stop services
docker-compose down

Model Cache: Models are cached in ~/.cache/huggingface and automatically mounted as a volume for persistence.

🎯 API Endpoints

POST /v1/audio/speech - Generate speech (OpenAI-compatible)
GET /v1/models - List available models
GET /v1/voices - List available voices
GET /health - Health check with backend status
GET /docs - Swagger UI documentation
GET /redoc - ReDoc documentation
GET /health - Health check endpoint
GET / - Web interface

🙏 Acknowledgments

This project builds upon the incredible work of the Qwen Team at Alibaba Cloud. We are deeply grateful for their development and open-sourcing of Qwen3-TTS, a state-of-the-art text-to-speech model that enables:

Powerful Speech Representation via Qwen3-TTS-Tokenizer-12Hz
Universal End-to-End Architecture with discrete multi-codebook LM
Extreme Low-Latency Streaming (as low as 97ms)
Intelligent Voice Control through natural language instructions

For more details about the underlying Qwen3-TTS models, please refer to:

News

2026.1.22: 🎉🎉🎉 We have released Qwen3-TTS series (0.6B/1.7B) based on Qwen3-TTS-Tokenizer-12Hz. Please check our blog!

Overview

Introduction

Qwen3-TTS covers 10 major languages (Chinese, English, Japanese, Korean, German, French, Russian, Portuguese, Spanish, and Italian) as well as multiple dialectal voice profiles to meet global application needs. In addition, the models feature strong contextual understanding, enabling adaptive control of tone, speaking rate, and emotional expression based on instructions and text semantics, and they show markedly improved robustness to noisy input text. Key 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.

Model Architecture

Released Models Description and Download

Below is an introduction and download information for the Qwen3-TTS models that have already been released. Other models mentioned in the technical report will be released in the near future. Please select and download the model that fits your needs.

Tokenizer Name	Description
Qwen3-TTS-Tokenizer-12Hz	The Qwen3-TTS-Tokenizer-12Hz model which can encode the input speech into codes and decode them back into speech.

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	✅
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	✅

During model loading in the qwen-tts package or vLLM, model weights will be automatically downloaded based on the model name. However, if your runtime environment is not conducive to downloading weights during execution, you can refer to the following commands to manually download the model weights to a local directory:

# Download through ModelScope (recommended for users in Mainland China)
pip install -U modelscope
modelscope download --model Qwen/Qwen3-TTS-Tokenizer-12Hz  --local_dir ./Qwen3-TTS-Tokenizer-12Hz 
modelscope download --model Qwen/Qwen3-TTS-12Hz-1.7B-CustomVoice --local_dir ./Qwen3-TTS-12Hz-1.7B-CustomVoice
modelscope download --model Qwen/Qwen3-TTS-12Hz-1.7B-VoiceDesign --local_dir ./Qwen3-TTS-12Hz-1.7B-VoiceDesign
modelscope download --model Qwen/Qwen3-TTS-12Hz-1.7B-Base --local_dir ./Qwen3-TTS-12Hz-1.7B-Base
modelscope download --model Qwen/Qwen3-TTS-12Hz-0.6B-CustomVoice --local_dir ./Qwen3-TTS-12Hz-0.6B-CustomVoice
modelscope download --model Qwen/Qwen3-TTS-12Hz-0.6B-Base --local_dir ./Qwen3-TTS-12Hz-0.6B-Base

# Download through Hugging Face
pip install -U "huggingface_hub[cli]"
huggingface-cli download Qwen/Qwen3-TTS-Tokenizer-12Hz --local-dir ./Qwen3-TTS-Tokenizer-12Hz
huggingface-cli download Qwen/Qwen3-TTS-12Hz-1.7B-CustomVoice --local-dir ./Qwen3-TTS-12Hz-1.7B-CustomVoice
huggingface-cli download Qwen/Qwen3-TTS-12Hz-1.7B-VoiceDesign --local-dir ./Qwen3-TTS-12Hz-1.7B-VoiceDesign
huggingface-cli download Qwen/Qwen3-TTS-12Hz-1.7B-Base --local-dir ./Qwen3-TTS-12Hz-1.7B-Base
huggingface-cli download Qwen/Qwen3-TTS-12Hz-0.6B-CustomVoice --local-dir ./Qwen3-TTS-12Hz-0.6B-CustomVoice
huggingface-cli download Qwen/Qwen3-TTS-12Hz-0.6B-Base --local-dir ./Qwen3-TTS-12Hz-0.6B-Base

Quickstart (Python Package)

Environment Setup

The easiest way to quickly use Qwen3-TTS is to install the qwen-tts Python package from PyPI. This will pull in the required runtime dependencies and allow you to load any released Qwen3-TTS model. We recommend using a fresh, isolated environment to avoid dependency conflicts with existing packages. You can create a clean Python 3.12 environment like this:

conda create -n qwen3-tts python=3.12 -y
conda activate qwen3-tts

then run:

pip install -U qwen-tts

If you want to develop or modify the code locally, install from source in editable mode.

git clone https://github.com/QwenLM/Qwen3-TTS.git
cd Qwen3-TTS
pip install -e .

Additionally, we recommend using FlashAttention 2 to reduce GPU memory usage.

pip install -U flash-attn --no-build-isolation

If your machine has less than 96GB of RAM and lots of CPU cores, run:

MAX_JOBS=4 pip install -U flash-attn --no-build-isolation

Also, you should have hardware that is compatible with FlashAttention 2. Read more about it in the official documentation of the FlashAttention repository. FlashAttention 2 can only be used when a model is loaded in torch.float16 or torch.bfloat16.

Python Package Usage

After installation, you can import Qwen3TTSModel to run custom voice TTS, voice design, and voice clone. The model weights can be specified either as a Hugging Face model id (recommended) or as a local directory path you downloaded. For all the generate_* functions below, besides the parameters shown and explicitly documented, you can also pass generation kwargs supported by Hugging Face Transformers model.generate, e.g., max_new_tokens, top_p, etc.

Custom Voice Generate

For custom voice models (Qwen3-TTS-12Hz-1.7B/0.6B-CustomVoice), you just need to call generate_custom_voice, passing a single string or a batch list, along with language, speaker, and optional instruct. You can also call model.get_supported_speakers() and model.get_supported_languages() to see which speakers and languages the current model supports.

import torch
import soundfile as sf
from qwen_tts import Qwen3TTSModel

model = Qwen3TTSModel.from_pretrained(
    "Qwen/Qwen3-TTS-12Hz-1.7B-CustomVoice",
    device_map="cuda:0",
    dtype=torch.bfloat16,
    attn_implementation="flash_attention_2",
)

# single inference
wavs, sr = model.generate_custom_voice(
    text="其实我真的有发现，我是一个特别善于观察别人情绪的人。",
    language="Chinese", # Pass `Auto` (or omit) for auto language adaptive; if the target language is known, set it explicitly.
    speaker="Vivian",
    instruct="用特别愤怒的语气说", # Omit if not needed.
)
sf.write("output_custom_voice.wav", wavs[0], sr)

# batch inference
wavs, sr = model.generate_custom_voice(
    text=[
        "其实我真的有发现，我是一个特别善于观察别人情绪的人。", 
        "She said she would be here by noon."
    ],
    language=["Chinese", "English"],
    speaker=["Vivian", "Ryan"],
    instruct=["", "Very happy."]
)
sf.write("output_custom_voice_1.wav", wavs[0], sr)
sf.write("output_custom_voice_2.wav", wavs[1], sr)

For Qwen3-TTS-12Hz-1.7B/0.6B-CustomVoice models, the supported speaker list and speaker descriptions are provided below. We recommend using each speaker’s native language for the best quality. Of course, each speaker can speak any language supported by the model.

Speaker	Voice Description	Native language
Vivian	Bright, slightly edgy young female voice.	Chinese
Serena	Warm, gentle young female voice.	Chinese
Uncle_Fu	Seasoned male voice with a low, mellow timbre.	Chinese
Dylan	Youthful Beijing male voice with a clear, natural timbre.	Chinese (Beijing Dialect)
Eric	Lively Chengdu male voice with a slightly husky brightness.	Chinese (Sichuan Dialect)
Ryan	Dynamic male voice with strong rhythmic drive.	English
Aiden	Sunny American male voice with a clear midrange.	English
Ono_Anna	Playful Japanese female voice with a light, nimble timbre.	Japanese
Sohee	Warm Korean female voice with rich emotion.	Korean

Voice Design

For the voice design model (Qwen3-TTS-12Hz-1.7B-VoiceDesign), you can use generate_voice_design to provide the target text and a natural-language instruct description.

import torch
import soundfile as sf
from qwen_tts import Qwen3TTSModel

model = Qwen3TTSModel.from_pretrained(
    "Qwen/Qwen3-TTS-12Hz-1.7B-VoiceDesign",
    device_map="cuda:0",
    dtype=torch.bfloat16,
    attn_implementation="flash_attention_2",
)

# single inference
wavs, sr = model.generate_voice_design(
    text="哥哥，你回来啦，人家等了你好久好久了，要抱抱！",
    language="Chinese",
    instruct="体现撒娇稚嫩的萝莉女声，音调偏高且起伏明显，营造出黏人、做作又刻意卖萌的听觉效果。",
)
sf.write("output_voice_design.wav", wavs[0], sr)

# batch inference
wavs, sr = model.generate_voice_design(
    text=[
      "哥哥，你回来啦，人家等了你好久好久了，要抱抱！",
      "It's in the top drawer... wait, it's empty? No way, that's impossible! I'm sure I put it there!"
    ],
    language=["Chinese", "English"],
    instruct=[
      "体现撒娇稚嫩的萝莉女声，音调偏高且起伏明显，营造出黏人、做作又刻意卖萌的听觉效果。",
      "Speak in an incredulous tone, but with a hint of panic beginning to creep into your voice."
    ]
)
sf.write("output_voice_design_1.wav", wavs[0], sr)
sf.write("output_voice_design_2.wav", wavs[1], sr)

Voice Clone

For the voice clone model (Qwen3-TTS-12Hz-1.7B/0.6B-Base), to clone a voice and synthesize new content, you just need to provide a reference audio clip (ref_audio) along with its transcript (ref_text). ref_audio can be a local file path, a URL, a base64 string, or a (numpy_array, sample_rate) tuple. If you set x_vector_only_mode=True, only the speaker embedding is used so ref_text is not required, but cloning quality may be reduced.

import torch
import soundfile as sf
from qwen_tts import Qwen3TTSModel

model = Qwen3TTSModel.from_pretrained(
    "Qwen/Qwen3-TTS-12Hz-1.7B-Base",
    device_map="cuda:0",
    dtype=torch.bfloat16,
    attn_implementation="flash_attention_2",
)

ref_audio = "https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen3-TTS-Repo/clone.wav"
ref_text  = "Okay. Yeah. I resent you. I love you. I respect you. But you know what? You blew it! And thanks to you."

wavs, sr = model.generate_voice_clone(
    text="I am solving the equation: x = [-b ± √(b²-4ac)] / 2a? Nobody can — it's a disaster (◍•͈⌔•͈◍), very sad!",
    language="English",
    ref_audio=ref_audio,
    ref_text=ref_text,
)
sf.write("output_voice_clone.wav", wavs[0], sr)

If you need to reuse the same reference prompt across multiple generations (to avoid recomputing prompt features), build it once with create_voice_clone_prompt and pass it via voice_clone_prompt.

prompt_items = model.create_voice_clone_prompt(
    ref_audio=ref_audio,
    ref_text=ref_text,
    x_vector_only_mode=False,
)
wavs, sr = model.generate_voice_clone(
    text=["Sentence A.", "Sentence B."],
    language=["English", "English"],
    voice_clone_prompt=prompt_items,
)
sf.write("output_voice_clone_1.wav", wavs[0], sr)
sf.write("output_voice_clone_2.wav", wavs[1], sr)

For more examples of reusable voice clone prompts, batch cloning, and batch inference, please refer to the example codes. With those examples and the generate_voice_clone function description, you can explore more advanced usage patterns.

Voice Design then Clone

If you want a designed voice that you can reuse like a cloned speaker, a practical workflow is: (1) use the VoiceDesign model to synthesize a short reference clip that matches your target persona, (2) feed that clip into create_voice_clone_prompt to build a reusable prompt, and then (3) call generate_voice_clone with voice_clone_prompt to generate new content without re-extracting features every time. This is especially useful when you want a consistent character voice across many lines.

import torch
import soundfile as sf
from qwen_tts import Qwen3TTSModel

# create a reference audio in the target style using the VoiceDesign model
design_model = Qwen3TTSModel.from_pretrained(
    "Qwen/Qwen3-TTS-12Hz-1.7B-VoiceDesign",
    device_map="cuda:0",
    dtype=torch.bfloat16,
    attn_implementation="flash_attention_2",
)

ref_text = "H-hey! You dropped your... uh... calculus notebook? I mean, I think it's yours? Maybe?"
ref_instruct = "Male, 17 years old, tenor range, gaining confidence - deeper breath support now, though vowels still tighten when nervous"
ref_wavs, sr = design_model.generate_voice_design(
    text=ref_text,
    language="English",
    instruct=ref_instruct
)
sf.write("voice_design_reference.wav", ref_wavs[0], sr)

# build a reusable clone prompt from the voice design reference
clone_model = Qwen3TTSModel.from_pretrained(
    "Qwen/Qwen3-TTS-12Hz-1.7B-Base",
    device_map="cuda:0",
    dtype=torch.bfloat16,
    attn_implementation="flash_attention_2",
)

voice_clone_prompt = clone_model.create_voice_clone_prompt(
    ref_audio=(ref_wavs[0], sr),   # or "voice_design_reference.wav"
    ref_text=ref_text,
)

sentences = [
    "No problem! I actually... kinda finished those already? If you want to compare answers or something...",
    "What? No! I mean yes but not like... I just think you're... your titration technique is really precise!",
]

# reuse it for multiple single calls
wavs, sr = clone_model.generate_voice_clone(
    text=sentences[0],
    language="English",
    voice_clone_prompt=voice_clone_prompt,
)
sf.write("clone_single_1.wav", wavs[0], sr)

wavs, sr = clone_model.generate_voice_clone(
    text=sentences[1],
    language="English",
    voice_clone_prompt=voice_clone_prompt,
)
sf.write("clone_single_2.wav", wavs[0], sr)

# or batch generate in one call
wavs, sr = clone_model.generate_voice_clone(
    text=sentences,
    language=["English", "English"],
    voice_clone_prompt=voice_clone_prompt,
)
for i, w in enumerate(wavs):
    sf.write(f"clone_batch_{i}.wav", w, sr)

Tokenizer Encode and Decode

If you only want to encode and decode audio for transport or training and so on, Qwen3TTSTokenizer supports encode/decode with paths, URLs, numpy waveforms, and dict/list payloads, for example:

import soundfile as sf
from qwen_tts import Qwen3TTSTokenizer

tokenizer = Qwen3TTSTokenizer.from_pretrained(
    "Qwen/Qwen3-TTS-Tokenizer-12Hz",
    device_map="cuda:0",
)

enc = tokenizer.encode("https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen3-TTS-Repo/tokenizer_demo_1.wav")
wavs, sr = tokenizer.decode(enc)
sf.write("decode_output.wav", wavs[0], sr)

For more tokenizer examples (including different input formats and batch usage), please refer to the example codes. With those examples and the description for Qwen3TTSTokenizer, you can explore more advanced usage patterns.

Launch Local Web UI Demo

To launch the Qwen3-TTS web ui demo, simply install the qwen-tts package and run qwen-tts-demo. Use the command below for help:

qwen-tts-demo --help

To launch the demo, you can use the following commands:

# CustomVoice model
qwen-tts-demo Qwen/Qwen3-TTS-12Hz-1.7B-CustomVoice --ip 0.0.0.0 --port 8000
# VoiceDesign model
qwen-tts-demo Qwen/Qwen3-TTS-12Hz-1.7B-VoiceDesign --ip 0.0.0.0 --port 8000
# Base model
qwen-tts-demo Qwen/Qwen3-TTS-12Hz-1.7B-Base --ip 0.0.0.0 --port 8000

And then open http://<your-ip>:8000, or access it via port forwarding in tools like VS Code.

Base Model HTTPS Notes

To avoid browser microphone permission issues after deploying the server, for Base model deployments, it is recommended/required to run the gradio service over HTTPS (especially when accessed remotely or behind modern browsers/gateways). Use --ssl-certfile and --ssl-keyfile to enable HTTPS. First we need to generate a private key and a self-signed cert (valid for 365 days):

openssl req -x509 -newkey rsa:2048 \
  -keyout key.pem -out cert.pem \
  -days 365 -nodes \
  -subj "/CN=localhost"

Then run the demo with HTTPS:

qwen-tts-demo Qwen/Qwen3-TTS-12Hz-1.7B-Base \
  --ip 0.0.0.0 --port 8000 \
  --ssl-certfile cert.pem \
  --ssl-keyfile key.pem \
  --no-ssl-verify

And open https://<your-ip>:8000 to experience it. If your browser shows a warning, it’s expected for self-signed certificates. For production, use a real certificate.

DashScope API Usage

To further explore Qwen3-TTS, we encourage you to try our DashScope API for a faster and more efficient experience. For detailed API information and documentation, please refer to the following:

API Description	API Documentation (Mainland China)	API Documentation (International)
Real-time API for Qwen3-TTS of custom voice model.	https://help.aliyun.com/zh/model-studio/qwen-tts-realtime	https://www.alibabacloud.com/help/en/model-studio/qwen-tts-realtime
Real-time API for Qwen3-TTS of voice clone model.	https://help.aliyun.com/zh/model-studio/qwen-tts-voice-cloning	https://www.alibabacloud.com/help/en/model-studio/qwen-tts-voice-cloning
Real-time API for Qwen3-TTS of voice design model.	https://help.aliyun.com/zh/model-studio/qwen-tts-voice-design	https://www.alibabacloud.com/help/en/model-studio/qwen-tts-voice-design

vLLM Usage

vLLM officially provides day-0 support for Qwen3-TTS! Welcome to use vLLM-Omni for Qwen3-TTS deployment and inference. For installation and more details, please check vLLM-Omni official documentation. Now only offline inference is supported. Online serving will be supported later, and vLLM-Omni will continue to offer support and optimization for Qwen3-TTS in areas such as inference speed and streaming capabilities.

Offline Inference

You can use vLLM-Omni to inference Qwen3-TTS locally, we provide examples in vLLM-Omni repo which can generate audio output:

# git clone https://github.com/vllm-project/vllm-omni.git

# cd vllm-omni/examples/offline_inference/qwen3_tts

# Run a single sample with CustomVoice task
python end2end.py --query-type CustomVoice

# Batch sample (multiple prompts in one run) with CustomVoice task:
python end2end.py --query-type CustomVoice --use-batch-sample

# Run a single sample with VoiceDesign task
python end2end.py --query-type VoiceDesign

# Batch sample (multiple prompts in one run) with VoiceDesign task:
python end2end.py --query-type VoiceDesign --use-batch-sample

# Run a single sample with Base task in icl mode-tag
python end2end.py --query-type Base --mode-tag icl

Fine Tuning

Please refer to the Qwen3-TTS Finetuning documentation in the upstream repository for detailed instructions on fine-tuning Qwen3-TTS.

Evaluation

During evaluation, we ran inference for all models with dtype=torch.bfloat16 and set max_new_tokens=2048. All other sampling parameters used the defaults from the checkpoint’s generate_config.json. For the Seed-Test and InstructTTS-Eval test sets, we set language="auto", while for all other test sets we explicitly passed the corresponding language. The detailed results are shown below.

Speech Generation Benchmarks

Zero-shot speech generation on the Seed-TTS test set. Performance is measured by Word Error Rate (WER, ↓), where lower is better.

Datasets	Model	Performance
Content Consistency
SEED test-zh \| test-en	Seed-TTS (Anastassiou et al., 2024)	1.12	2.25
	MaskGCT (Wang et al., 2024)	2.27	2.62
	E2 TTS (Eskimez et al., 2024)	1.97	2.19
	F5-TTS (Chen et al., 2024)	1.56	1.83
	Spark TTS (Wang et al., 2025)	1.20	1.98
	Llasa-8B (Ye et al., 2025b)	1.59	2.97
	KALL-E (Xia et al., 2024)	0.96	1.94
	FireRedTTS 2 (Xie et al., 2025)	1.14	1.95
	CosyVoice 3 (Du et al., 2025)	0.71	1.45
	MiniMax-Speech (Zhang et al., 2025a)	0.83	1.65
	Qwen3-TTS-25Hz-0.6B-Base	1.18	1.64
	Qwen3-TTS-25Hz-1.7B-Base	1.10	1.49
	Qwen3-TTS-12Hz-0.6B-Base	0.92	1.32
	Qwen3-TTS-12Hz-1.7B-Base	0.77	1.24

Multilingual speech generation on the TTS multilingual test set. Performance is measured by Word Error Rate (WER, ↓) for content consistency and Cosine Similarity (SIM, ↑) for speaker similarity.

Language	Qwen3-TTS-25Hz		Qwen3-TTS-12Hz		MiniMax	ElevenLabs
Language	0.6B-Base	1.7B-Base	0.6B-Base	1.7B-Base	MiniMax	ElevenLabs
Content Consistency
Chinese	1.108	0.777	1.145	0.928	2.252	16.026
English	1.048	1.014	0.836	0.934	2.164	2.339
German	1.501	0.960	1.089	1.235	1.906	0.572
Italian	1.169	1.105	1.534	0.948	1.543	1.743
Portuguese	2.046	1.778	2.254	1.526	1.877	1.331
Spanish	2.031	1.491	1.491	1.126	1.029	1.084
Japanese	4.189	5.121	6.404	3.823	3.519	10.646
Korean	2.852	2.631	1.741	1.755	1.747	1.865
French	2.852	2.631	2.931	2.858	4.099	5.216
Russian	5.957	4.535	4.458	3.212	4.281	3.878
Speaker Similarity
Chinese	0.797	0.796	0.811	0.799	0.780	0.677
English	0.811	0.815	0.829	0.775	0.756	0.613
German	0.749	0.737	0.769	0.775	0.733	0.614
Italian	0.722	0.718	0.792	0.817	0.699	0.579
Portuguese	0.790	0.783	0.794	0.817	0.805	0.711
Spanish	0.732	0.731	0.812	0.814	0.762	0.615
Japanese	0.810	0.807	0.798	0.788	0.776	0.738
Korean	0.824	0.814	0.812	0.799	0.779	0.700
French	0.698	0.703	0.700	0.714	0.628	0.535
Russian	0.734	0.744	0.781	0.792	0.761	0.676

Cross-lingual speech generation on the Cross-Lingual benchmark. Performance is measured by Mixed Error Rate (WER for English, CER for others, ↓).

Task	Qwen3-TTS-25Hz-1.7B-Base	Qwen3-TTS-12Hz-1.7B-Base	CosyVoice3	CosyVoice2
en-to-zh	5.66	4.77	5.09	13.5
ja-to-zh	3.92	3.43	3.05	48.1
ko-to-zh	1.14	1.08	1.06	7.70
zh-to-en	2.91	2.77	2.98	6.47
ja-to-en	3.95	3.04	4.20	17.1
ko-to-en	3.48	3.09	4.19	11.2
zh-to-ja	9.29	8.40	7.08	13.1
en-to-ja	7.74	7.21	6.80	14.9
ko-to-ja	4.17	3.67	3.93	5.86
zh-to-ko	8.12	4.82	14.4	24.8
en-to-ko	6.83	5.14	5.87	21.9
ja-to-ko	6.86	5.59	7.92	21.5

Controllable speech generation on InstructTTSEval. Performance is measured by Attribute Perception and Synthesis accuracy (APS), Description-Speech Consistency (DSD), and Response Precision (RP).

Type	Model	InstructTTSEval-ZH			InstructTTSEval-EN
Type	Model	APS (↑)	DSD (↑)	RP (↑)	APS (↑)	DSD (↑)	RP (↑)
Target Speaker	Gemini-flash	88.2	90.9	77.3	92.3	93.8	80.1
	Gemini-pro	89.0	90.1	75.5	87.6	86.0	67.2
	Qwen3TTS-25Hz-1.7B-CustomVoice	83.1	75.0	63.0	79.0	82.8	69.3
	Qwen3TTS-12Hz-1.7B-CustomVoice	83.0	77.8	61.2	77.3	77.1	63.7
	GPT-4o-mini-tts	54.9	52.3	46.0	76.4	74.3	54.8
Voice Design	Qwen3TTS-12Hz-1.7B-VD	85.2	81.1	65.1	82.9	82.4	68.4
	Mimo-Audio-7B-Instruct (Zhang et al., 2025b)	75.7	74.3	61.5	80.6	77.6	59.5
	VoiceSculptor (Hu et al., 2026)	75.7	64.7	61.5	-	-	-
	Hume	-	-	-	83.0	75.3	54.3
	VoxInstruct (Zhou et al., 2024)	47.5	52.3	42.6	54.9	57.0	39.3
	Parler-tts-mini (Lyth & King, 2024)	-	-	-	63.4	48.7	28.6
	Parler-tts-large (Lyth & King, 2024)	-	-	-	60.0	45.9	31.2
	PromptTTS (Guo et al., 2023)	-	-	-	64.3	47.2	31.4
	PromptStyle (Liu et al., 2023)	-	-	-	57.4	46.4	30.9

Target-Speaker Multilingual Speech Generation on the TTS multilingual test set. Performance is measured by Word Error Rate (WER, ↓).

Language	Qwen3-TTS-25Hz		Qwen3-TTS-12Hz		GPT-4o-Audio Preview
Language	0.6B-CustomVoice	1.7B-CustomVoice	0.6B-CustomVoice	1.7B-CustomVoice	GPT-4o-Audio Preview
Chinese	0.874	0.708	0.944	0.903	3.519
English	1.332	0.936	1.188	0.899	2.197
German	0.990	0.634	2.722	1.057	1.161
Italian	1.861	1.271	2.545	1.362	1.194
Portuguese	1.728	1.854	3.219	2.681	1.504
Spanish	1.309	1.284	1.154	1.330	4.000
Japanese	3.875	4.518	6.877	4.924	5.001
Korean	2.202	2.274	3.053	1.741	2.763
French	3.865	3.080	3.841	3.781	3.605
Russian	6.529	4.444	5.809	4.734	5.250

Long speech generation results. Performance is measured by Word Error Rate (WER, ↓).

Datasets	Model	Performance
Content Consistency
long-zh \| long-en	Higgs-Audio-v2 (chunk) (Boson AI, 2025)	5.505	6.917
	VibeVoice (Peng et al., 2025)	22.619	1.780
	VoxCPM (Zhou et al., 2025)	4.835	7.474
	Qwen3-TTS-25Hz-1.7B-CustomVoice	1.517	1.225
	Qwen3-TTS-12Hz-1.7B-CustomVoice	2.356	2.812

Speech Tokenizer Benchmarks

Comparison between different supervised semantic speech tokenizers on ASR Task.

Model	Codebook Size	FPS	C.V. EN	C.V. CN	Fluers EN	Fluers CN
S3 Tokenizer(VQ) (Du et al., 2024a)	4096	50	12.06	15.38	-	-
S3 Tokenizer(VQ) (Du et al., 2024a)	4096	25	11.56	18.26	7.65	5.03
S3 Tokenizer(FSQ) (Du et al., 2024a)	6561	25	10.67	7.29	6.58	4.43
Qwen-TTS-Tokenizer-25Hz (Stage 1)	32768	25	7.51	10.73	3.07	4.23
Qwen-TTS-Tokenizer-25Hz (Stage 2)	32768	25	10.40	14.99	4.14	4.67

Comparison between different semantic-related speech tokenizers.

Model	NQ	Codebook Size	FPS	PESQ_WB	PESQ_NB	STOI	UTMOS	SIM
SpeechTokenizer (Zhang et al., 2023a)	8	1024	50	2.60	3.05	0.92	3.90	0.85
X-codec (Ye et al., 2025a)	2	1024	50	2.68	3.27	0.86	4.11	0.84
X-codec 2 (Ye et al., 2025b)	1	65536	50	2.43	3.04	0.92	4.13	0.82
XY-Tokenizer (Gong et al., 2025)	8	1024	12.5	2.41	3.00	0.91	3.98	0.83
Mimi (Défossez et al., 2024)	16	2048	12.5	2.88	3.42	0.94	3.87	0.87
FireredTTS 2 Tokenizer (Xie et al., 2025)	16	2048	12.5	2.73	3.28	0.94	3.88	0.87
Qwen-TTS-Tokenizer-12Hz	16	2048	12.5	3.21	3.68	0.96	4.16	0.95

Citation

If you find our paper and code useful in your research, please consider giving a star ⭐ and citation 📝 :)

@article{Qwen3-TTS,
  title={Qwen3-TTS Technical Report},
  author={Hangrui Hu and Xinfa Zhu and Ting He and Dake Guo and Bin Zhang and Xiong Wang and Zhifang Guo and Ziyue Jiang and Hongkun Hao and Zishan Guo and Xinyu Zhang and Pei Zhang and Baosong Yang and Jin Xu and Jingren Zhou and Junyang Lin},
  journal={arXiv preprint arXiv:2601.15621},
  year={2026}
}

Name		Name	Last commit message	Last commit date
Latest commit History 48 Commits
.github		.github
api		api
docs		docs
tasks		tasks
tests		tests
.gitignore		.gitignore
BENCHMARK_RESULTS.md		BENCHMARK_RESULTS.md
DEPLOYMENT_INVESTIGATION.md		DEPLOYMENT_INVESTIGATION.md
DEPLOYMENT_STATUS.md		DEPLOYMENT_STATUS.md
Dockerfile		Dockerfile
Dockerfile.vllm		Dockerfile.vllm
LICENSE		LICENSE
MANIFEST.in		MANIFEST.in
OPTIMIZATION_GUIDE.md		OPTIMIZATION_GUIDE.md
OPTIMIZATION_SUMMARY.md		OPTIMIZATION_SUMMARY.md
README.md		README.md
UPDATE_NOTES.md		UPDATE_NOTES.md
VLLM_BACKEND_STATUS.md		VLLM_BACKEND_STATUS.md
VOICE_CLONING_PERSISTENCE.md		VOICE_CLONING_PERSISTENCE.md
bench_tts.py		bench_tts.py
benchmark_official.py		benchmark_official.py
docker-compose.yml		docker-compose.yml
extended_warmup.py		extended_warmup.py
optimization_verification.txt		optimization_verification.txt
pyproject.toml		pyproject.toml
pytest.ini		pytest.ini
start_server.sh		start_server.sh
test_optimizations.py		test_optimizations.py
test_opts_simple.py		test_opts_simple.py
verify_optimizations.py		verify_optimizations.py

Folders and files

Latest commit

History

Repository files navigation

Qwen3-TTS

✨ Features

Backend Options

🚀 Performance Benchmarks

Flash Attention 2 Impact

Detailed Results (with Flash Attention 2)

⚡ Performance Optimizations

🚀 Quick Start (API Server)

Using OpenAI Python Client

Language-Specific Models

Web Interface

📦 Deployment

Option 1: Using Conda (Recommended for Development)

Option 2: Using Docker (NVIDIA GPU)

Docker Compose Configuration

🎯 API Endpoints

🙏 Acknowledgments

News

Contents

Overview

Introduction

Model Architecture

Released Models Description and Download

Quickstart (Python Package)

Environment Setup

Python Package Usage

Custom Voice Generate

Voice Design

Voice Clone

Voice Design then Clone

Tokenizer Encode and Decode

Launch Local Web UI Demo

Base Model HTTPS Notes

DashScope API Usage

vLLM Usage

Offline Inference

Fine Tuning

Evaluation

Citation

Star History

About

Resources

License

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

Packages