anke01/qwen3-tts-uyghur

Qwen3-TTS Uyghur Fine-tuned Model

维吾尔语 TTS 微调模型，基于 Qwen3-TTS-12Hz-0.6B-Base 微调。

模型信息

项目	值
基础模型	Qwen3-TTS-12Hz-0.6B-Base
训练数据	Common Voice Uyghur v24 (14,211 样本)
说话人	uyghur_speaker
采样率	24000 Hz
训练步数	4000 steps
最终 Loss	~4.5

使用示例

Python

from qwen_tts import Qwen3TTSModel
import soundfile as sf
import torch

# 加载模型
tts = Qwen3TTSModel.from_pretrained(
    "anke01/qwen3-tts-uyghur",
    device_map="cuda:0",
    dtype=torch.bfloat16,
)

# 合成语音
wavs, sr = tts.generate_custom_voice(
    text="بۇ بىر سىناق ئاۋازى.",
    speaker="uyghur_speaker",
    language="Auto",
)

# 保存
sf.write("output.wav", wavs[0], sr)
print(f"Saved: {len(wavs[0])/sr:.2f}s at {sr}Hz")

命令行

# 下载模型
git lfs install
git clone https://huggingface.co/anke01/qwen3-tts-uyghur

# 推理
cd qwen3-tts-uyghur
python infer.py \
    --model_path ./checkpoint-step-4000 \
    --text "بۇ بىر سىناق ئاۋازى." \
    --speaker uyghur_speaker \
    --output test.wav

训练配置

参数	值
Batch Size	2
Gradient Accumulation	16
Learning Rate	2e-6
Max Steps	4000
Epoch	0 (55%)

文件结构

qwen3-tts-uyghur/
├── checkpoint-step-4000/    # 微调模型权重
│   ├── config.json
│   ├── model.safetensors    # 1.8GB
│   ├── tokenizer_config.json
│   ├── vocab.json
│   ├── merges.txt
│   └── speech_tokenizer/
├── train.py                 # 训练脚本
├── infer.py                 # 推理脚本
├── dataset.py               # 数据集类
├── prepare_data.py          # 数据预处理
├── convert_commonvoice_to_jsonl.py
├── requirements.txt
└── README.md

训练数据格式

{
  "audio": "path/to/audio.wav",
  "text": "维吾尔语文本",
  "ref_audio": "path/to/reference.wav",
  "audio_codes": [[...], [...]]
}

依赖安装

pip install -r requirements.txt

requirements.txt:

qwen-tts==0.1.1
transformers>=4.57.3
accelerate>=0.25.0
torch>=2.0.0
torchaudio>=2.0.0
librosa>=0.10.0
soundfile>=0.12.0
tqdm>=4.66.0
safetensors>=0.4.0

示例文本

بۇ بىر سىناق ئاۋازى.           # 这是一个测试声音
ياخشىمۇسىز، بۈگۈن ھاۋا ياخشى.  # 你好，今天天气好
ئىككىنچى سىناق.                # 第二个测试

许可证

Apache 2.0

相关链接

• Qwen3-TTS GitHub
• Qwen3-TTS-12Hz-0.6B-Base
• Common Voice Uyghur
• Qwen Chat

引用

@misc{qwen3tts2025,
  title={Qwen3-TTS: A Multilingual Multimodal TTS Model},
  author={Qwen Team},
  year={2025},
  url={https://github.com/QwenLM/Qwen3-TTS}
}