Model · Alibaba · Text-To-Speech
Qwen3-TTS
Alibaba's open text-to-speech family. The headline 1.7B CustomVoice model speaks 10 languages in nine preset voices you can steer with plain-language instructions; sibling models clone a voice from 3 seconds of audio or design one from a description.
- Modality
- Text-To-Speech
- License
- Apache 2.0 (Open)
- Parameter size
- 1.7B (headline); 0.6B siblings
- Released
- January 22, 2026
- Last verified
- October 28, 2026
- Runs locally
- Yes
What you need to run it
- Mac
- Yes. Any Apple Silicon Mac (M1 or newer) with at least 8 GB (16 GB recommended) of memory. Newer chips are faster, but memory is what decides whether it runs.
- Windows / Linux PC
- NVIDIA works best. A graphics card with at least 6 GB (8 GB recommended) of video memory.
- No graphics card
- Not practical. Too slow without a graphics card or Apple Silicon.
MLX builds from mlx-community run it on a Mac; the official package targets CUDA graphics cards.
Numbers are for: 0.6B for the minimum, 1.7B (bf16) for the recommended. Our estimate from the model’s size — the maker publishes no figure.
Strengths
- Apache 2.0 — no commercial restrictions
- Clones a voice from a 3-second sample (the Base models)
- Voice Design from a text description (1.7B VoiceDesign model)
- Instruction control over tone, emotion and pace in plain language
- Streaming with first audio after a single character — end-to-end latency as low as 97 ms
- Small enough to self-host: the largest model is 1.7B parameters
Weaknesses
- 10 languages (Chinese, English, Japanese, Korean, German, French, Russian, Portuguese, Spanish, Italian) — fewer than VoxCPM2 or OmniVoice
- Voice Design and instruction control exist only at 1.7B; the 0.6B models do not have them
- Five separate checkpoints — pick the right one for clone, design or preset voices
Try it
| Where | Type | Notes |
|---|---|---|
| Hugging Face | weights | Apache 2.0; 1.7B CustomVoice, VoiceDesign and Base, plus 0.6B CustomVoice and Base |
| GitHub (qwen-tts package) | local | Python package, local web UI demo, vLLM support |
| Alibaba Cloud Model Studio (DashScope) | hosted-api | API key required |
Official sources
- Model card (1.7B CustomVoice) model
- GitHub github
- Qwen blog announcement
- Paper paper
Change log
- — Added hardware requirements (Mac support, memory, NVIDIA). Source: GitHub README + mlx-community/Qwen3-TTS-12Hz-1.7B-Base-8bit + size.
- — Initial entry. Added 2026-10-28 from the popularity sweep: 14.1k GitHub stars, 1.88M Hugging Face downloads/month for the 1.7B CustomVoice model. Released 2026-01-22 per the GitHub README news; no newer Qwen TTS release on Hugging Face.
Esc