KaniTTS-400M-KY-Kani (Ascend NPU 适配版)

pip install kani-tts
pip install -U "transformers==4.57.1"
pip install torch_npu  # 华为昇腾环境

# 方式1: 使用适配后的推理脚本
python inference.py --text "Салам, менин атым Акылай." --output output.wav

# 方式2: 编程接口
from kani_tts import KaniTTS

model = KaniTTS('/path/to/kani-tts-400m-ky-kani')
audio, text = model("Салам, менин атым Акылай.")
model.save_audio(audio, "output.wav")

# 关键适配代码
torch.cuda.is_available = lambda: True
torch.cuda.current_device = lambda: 0
torch.Tensor.cuda = lambda self: self.npu()
torch.nn.Module.cuda = lambda self: self.npu()

输入文本 → Tokenizer → LFM2 LLM (NPU) → 音频Tokens → NanoCodec (NPU) → 音频波形

kani-tts-400m-ky-kani/
├── model.safetensors       # 模型权重 (743MB)
├── config.json              # 模型配置
├── tokenizer.json           # 分词器
├── inference.py             # NPU 适配推理脚本
├── evaluate.py              # 性能/精度评测脚本
├── eval/                    # 评测结果
│   └── evaluation_report.json
├── README_NPU.md            # 本文件
└── README.md                # 原始 README

@inproceedings{emilialarge,
  author={He, Haorui and Shang, Zengqiang and Wang, Chaoren and Li, Xuyuan and Gu, Yicheng and Hua, Hua and Liu, Liwei and Yang, Chen and Li, Jiaqi and Shi, Peiyang and Wang, Yuancheng and Chen, Kai and Zhang, Pengyuan and Wu, Zhizheng},
  title={Emilia: A Large-Scale, Extensive, Multilingual, and Diverse Dataset for Speech Generation},
  booktitle={arXiv:2501.15907},
  year={2025}
}