Qwen3 TTS 1.7B

Qwen3 TTS 1.7B is Alibaba's state-of-the-art text-to-speech model with 1.87 million monthly downloads. It supports 10 languages with voice cloning, voice design, and ultra-low 97ms streaming latency. Apache 2.0 licensed.

Key features

1. 10 languages - Chinese, English, Japanese, Korean, German, French, Russian, Portuguese, Spanish, Italian
2. Voice cloning - 3-second reference audio to clone any voice
3. Voice design - natural language voice descriptions
4. 97ms latency - extreme low-latency streaming
5. 9 premium speakers - diverse built-in voices
6. vLLM support - production deployment via vLLM-Omni
7. 3 model variants - CustomVoice, VoiceDesign, Base (clone)

Model variants

Quick start

pip install -U qwen-tts

from qwen_tts import Qwen3TTSModel
import torch
import soundfile as sf
 
model = Qwen3TTSModel.from_pretrained(
    "Qwen/Qwen3-TTS-12Hz-1.7B-CustomVoice",
    device_map="cuda:0",
    dtype=torch.bfloat16,
)
 
wavs, sr = model.generate_custom_voice(
    text="Hello! This is Qwen3 TTS speaking.",
    language="English",
    speaker="Ryan",
)
sf.write("output.wav", wavs[0], sr)

When to use

• Voice assistants - low-latency streaming TTS
• Content creation - multilingual voiceovers
• Accessibility - screen readers and narration
• Audiobooks - emotional tone control
• Voice cloning - personalized voices