Qwen3-TTS-API
OpenAI-compatible Text-to-Speech API powered by Qwen3-TTS with CUDA graph acceleration, delivering 7-10x faster inference across 10 languages with 9 built-in…
Platform availability
Available on 1 of 10 platforms
- Unraid
- TrueNAS (not listed)
- Umbrel (not listed)
- ZimaOS (not listed)
- Proxmox (not listed)
- Helm (not listed)
- Mac (not listed)
- Windows (not listed)
- Linux (not listed)
- Any server (not listed)
Health score?A 0-100 score computed nightly from four factors: maintained (40%: recent commits and steady activity), popular (25%: GitHub stars, log scale), easy to install (20%: platforms, install notes, setup guides, Docker image) and light to run (15%: minimum RAM, ARM support). When the source code has been checked, a fifth factor, security check (20%: automated scan for committed secrets, unsafe desktop-app settings, known-vulnerable dependencies and project hygiene; any serious finding caps it at 20), is added. Factors without data are left out and the rest are rescaled.
41/100 Fair
- Maintained
- 59last commit 165 days ago
- Popular
- 123 GitHub stars
- Easy to install
- 351 install platform, install notes, Docker image
- Light to run
- 501.09961 GB minimum RAM
Availability: platform listings confirmed today, links checked 18 days ago
- GitHub stars
- 3
- Open issues
- 0
- Last commit
- 2026-06-09
- Activity
- stale
Checked 6 days ago - source: GitHub (hsiang-han/Qwen3-TTS-API)
Resources & Compatibility
Minimum requirements: 1.09961 GB RAM.
Checked 54 days ago - source: Docker Hub image tags and published docs
First-install notes
- Use GPU and map your files Install the NVIDIA Container Toolkit (the add-on that lets Docker use your graphics card), then replace /path/to/your/workspace with a real folder on your computer before running the Docker command. Without the toolkit, Docker can't use your GPU, and without the correct folder the container won't find your audio files. source
- Set a free web port Change HOST_PORT to a port that isn't already in use, because that is the port you open in your browser or send API requests to. If it is busy, the app may start but you won't be able to reach it. source
- Leave GPU memory headroom If you run the vLLM server (a fast AI engine), keep the GPU memory setting below 100%, such as the 0.8 example. Using too much can cause out-of-memory crashes, especially if other programs are also using your GPU. source
- Model Download & GPU Setup First start downloads a ~3-7 GB model from HuggingFace; allow that much free disk space before launching. Confirm your NVIDIA driver is 550 or newer. China-based users should set the HF_ENDPOINT environment variable to the HuggingFace mirror (hf-mirror.com) for faster model downloads. source
- NVIDIA GPU & Driver Requirement Requires an NVIDIA GPU (RTX 4060 or 5060 Ti class or newer) with driver 550+. No flash-attn, vLLM, or Triton needed — inference uses CUDA graphs only. source
Fetched 45 days ago - each note links to its source
Variants
Alternatives
No alternatives collected yet.
Common questions
How do I deploy Qwen3-TTS-API with Docker?
Run the container with GPU access, map port 8080, mount the model cache at /root/.cache/huggingface, and set --shm-size=4g. A Docker Compose file is also provided in the repo's docker/gpu directory.
Which endpoint do I call for standard text-to-speech?
Use the OpenAI-compatible /v1/audio/speech endpoint with a JSON body containing input and voice, with optional language, instruct, or stream parameters.
How do I use voice cloning?
Use the /v1/audio/speech/clone endpoint. The README says voice cloning is supported from 3-second reference audio.
How do I control emotion in the generated speech?
Emotion control is available only with the 1.7B CustomVoice model. Send an instruct field in the /v1/audio/speech request.
How should China-based users handle model downloads?
Set the HF_ENDPOINT environment variable to the HuggingFace mirror for China.
Answers sourced from github.com
Community
Collected 49 days ago - public community threads