ð£ïž Qwen3-TTS Docker â All-in-One
Production-ready Docker deployment for Qwen3-TTS with a modern Web UI and full REST API. All models baked into a single Docker image â no downloads at runtime.
ãšãŒãžã§ã³ãã匷å
å ¬éå ããã€ã³ã¹ããŒã«ããŸããPlugMCP ã®ãšã³ããã€ã³ãã§ã¯ãããŸããã
ããŒãšãã°ä»ãã§ã³ããŒããã¹ãã
ã¬ãžã¹ããªã«ããã±ãŒãžçã®ãªãªãŒã¹ã¯ãŸã ãããŸããã ãªããžããªã® README ãåç §.
ãªã¢ãŒã URL ããæå ã®ãã·ã³ã®ã³ãã³ãã 2 ã€ã®ã€ã³ã¹ããŒã«æ¹æ³ã®éã
PlugMCP ã§å®è¡
A new server you create. It is not this listing's official endpoint. Use the publisher install when you want their URL and their credentials.
èµ·åå ¬éå ãã
Production-ready Docker deployment for Qwen3-TTS with a modern Web UI and full REST API. All models baked into a single Docker image â no downloads at runtime.
ð£ïž Qwen3-TTS All-in-One Docker: Web UI + REST API + MCP Server | 10 Languages | Voice Clone/Design/Custom | GPU Auto-Management
/docs.pt file, reuse across sessions# Pull the all-in-one image (~30GB with models)
docker pull neosun/qwen3-tts:2.0.0
# Run with GPU
docker run -d --name qwen3-tts \
--gpus '"device=0"' \
-p 8766:8766 \
neosun/qwen3-tts:2.0.0
git clone https://github.com/neosun100/Qwen3-TTS.git
cd Qwen3-TTS
# Edit .env to set your GPU
echo "NVIDIA_VISIBLE_DEVICES=0" > .env
docker compose up -d
git clone https://github.com/neosun100/Qwen3-TTS.git
cd Qwen3-TTS
bash start.sh
Then open: http://localhost:8766
curl -X POST http://localhost:8766/api/tts/custom-voice \
-H "Content-Type: application/json" \
-d '{
"text": "Hello, this is a test.",
"language": "English",
"speaker": "Ryan",
"instruct": "Speak with excitement"
}' -o output.wav
curl -X POST http://localhost:8766/api/tts/voice-design \
-H "Content-Type: application/json" \
-d '{
"text": "Hello world!",
"language": "English",
"instruct": "Young female voice, cheerful and bright"
}' -o output.wav
curl -X POST http://localhost:8766/api/tts/voice-clone \
-F "text=Hello from a cloned voice" \
-F "language=English" \
-F "ref_text=This is the reference transcript" \
-F "[email protected]" \
-o output.wav
TTS endpoints return timing headers for performance monitoring:
Important: Qwen3-TTS does not support true incremental audio streaming at the model level. The official Qwen team confirmed this in Issue #10:
"Streaming inference is supported at the model architecture level. Currently, our qwen-tts package primarily focuses on enabling quick demos... For streaming capabilities, ongoing development will be mainly driven by the vLLM-Omni community."
The /stream endpoints in this project generate the full audio first, then deliver it as chunked PCM data. This enables progressive playback via Web Audio API but does not reduce time-to-first-byte (TTFB).
For true streaming, monitor:
If you want to build the Docker image yourself:
git clone https://github.com/neosun100/Qwen3-TTS.git
cd Qwen3-TTS
# Download models first (~14GB total)
pip install -U "huggingface_hub[cli]"
mkdir -p models
huggingface-cli download Qwen/Qwen3-TTS-Tokenizer-12Hz --local-dir ./models/Qwen3-TTS-Tokenizer-12Hz
huggingface-cli download Qwen/Qwen3-TTS-12Hz-1.7B-CustomVoice --local-dir ./models/Qwen3-TTS-12Hz-1.7B-CustomVoice
huggingface-cli download Qwen/Qwen3-TTS-12Hz-1.7B-VoiceDesign --local-dir ./models/Qwen3-TTS-12Hz-1.7B-VoiceDesign
huggingface-cli download Qwen/Qwen3-TTS-12Hz-1.7B-Base --local-dir ./models/Qwen3-TTS-12Hz-1.7B-Base
# Build (uses pre-compiled flash-attn wheel â fast!)
docker build -t neosun/qwen3-tts:2.0.0 .
Building flash-attn from source requires CUDA development headers and takes 40+ minutes even with ninja. This Dockerfile downloads a pre-compiled wheel matching Python 3.12 + CUDA 12 + PyTorch 2.9, reducing install time to seconds.
The wheel is from the official flash-attention releases.
This project is licensed under the Apache 2.0 License. The Qwen3-TTS models are subject to the Qwen License.
ð£ïž Qwen3-TTS All-in-One Docker: Web UI + REST API + MCP Server | 10 Languages | Voice Clone/Design/Custom | GPU Auto-Management
ããŒã«
ããŒã«äžèЧã¯ãŸã ãããŸããããªã¢ãŒããµãŒããŒã¯ãæåã®ãã«ã¹ãã§ãã¯åŸã«ååŸãããŸãã
ãããã質å
ãã®ããŒãžã§ CursorãClaude DesktopãClaude CodeãVS Code çšã® stdio èšå®ãã³ããŒã§ããŸããPlugMCP ã®ãããã·ã§ã¯ãªããå ¬éå ã®ã€ã³ã¹ããŒã«æ¹æ³ã䜿ããŸãã
ãã®ã¬ã³ãŒãã«å¿ é ã®ç°å¢å€æ°ã¯èšèŒãããŠããŸããã
Qwen3 Tts 㯠stdio çµç±ã§ããŒã«ã«å®è¡ãããŸãã
æçµãã§ãã¯ã¯ 2026-10-28 ã§ãã ð£ïž Qwen3-TTS All-in-One Docker: Web UI + REST API + MCP Server | 10 Languages | Voice Clone/Design/Custom | GPU Auto-Management