Install Qwen3-TTS on Windows with Conda

Introduction

Qwen3-TTS is an open-source series of TTS models developed by the Qwen team at Alibaba Cloud, supporting stable, expressive, and streaming speech generation, free-form voice design, and vivid voice cloning.

GitHub: https://github.com/QwenLM/Qwen3-TTS
Hugging Face: https://hf.co/collections/Qwen/qwen3-tts
Demo: https://hf.co/spaces?q=Qwen3-TTS

Video tutorial

Coming soon

Step 1. Install Miniconda Package

Download Miniconda: https://www.anaconda.com/download/success?reg=skipped

Direct link: https://anaconda.com/api/installers/Miniconda3-latest-Windows-x86_64.exe

Step 2. Create Conda Environment

Create a conda environment:

name: qwen3-tts
channels:
  - pytorch
  - nvidia
  - defaults
dependencies:
  - python=3.12
  - pip
  - pip:
      # PyTorch CUDA 12.6 wheels
      - --extra-index-url https://download.pytorch.org/whl/cu126
      - torch
      - torchvision
      - torchaudio

      # Qwen3-TTS
      - qwen-tts

      # Read/write audio files
      - soundfile

      # Download the model from Hugging Face / ModelScope
      - huggingface_hub[cli]
      - modelscope

Activate conda environment:

conda env create -f environment.yml
conda activate qwen3-tts

Step 3. Create a Python Script

Create a file named qwen3_tts_test.py:

import torch
import soundfile as sf
from qwen_tts import Qwen3TTSModel

# Use float16 instead of bfloat16 if your older GPU doesn't support bfloat16.
model = Qwen3TTSModel.from_pretrained(
    "Qwen/Qwen3-TTS-12Hz-1.7B-CustomVoice",
    device_map="cuda:0",
    dtype=torch.bfloat16,
    # Do not use attn_implementation="flash_attention_2" on Windows.
    #attn_implementation="flash_attention_2",
)

# single inference
wavs, sr = model.generate_custom_voice(
    text="Hello, this is a test.",
    language="English",
    speaker="Ryan",
)

sf.write("output.wav", wavs[0], sr)
print("output.wav has been saved!")

Run it:

python qwen3_tts_test.py

The result will be the audio file output.wav

Kokoro 82M Miniconda