README.md

---
library_name: transformers
pipeline_tag: audio-to-audio
tags:
- signal-processing
license: apache-2.0
---


<div align="center">
    <h1>
    Dasheng Denoiser
    </h1>
    <p>
    Official PyTorch inference code for the Interspeech 2025 paper: <br>
    <b><em>Efficient Speech Enhancement via Embeddings from Pre-trained Generative Audioencoders</em></b>
    </p>
    <a href="https://arxiv.org/abs/2506.11514"><img src="https://img.shields.io/badge/arxiv-2506.11514-red" alt="version"></a>
    <a href="https://www.python.org"><img src="https://img.shields.io/badge/Python-3.10+-orange" alt="version"></a>
    <a href="https://pytorch.org"><img src="https://img.shields.io/badge/PyTorch-2.0+-brightgreen" alt="python"></a>
    <a href="https://www.apache.org/licenses/LICENSE-2.0"><img src="https://img.shields.io/badge/License-Apache%202.0-blue.svg" alt="mit"></a>
    <a href="https://github.com/xiaomi-research/dasheng-denoiser"><img src="https://img.shields.io/github/stars/xiaomi-research/dasheng-denoiser?style=social" alt="stars"></a>


</div>


# Installation and Usage

```bash
uv pip install transformers torch torchaudio einops
```

```python
import torch
import torchaudio
from transformers import AutoModel
model = AutoModel.from_pretrained("mispeech/dasheng-denoiser",  trust_remote_code=True)
model.eval()
# Load audio file (only 16kHz supported!)
audio, sr = torchaudio.load("path/to/audio.wav")
with torch.no_grad(), torch.autocast(device_type='cuda'):
    enhanced = model(audio)
torchaudio.save("enhanced_audio.wav", enhanced, sr)
```


# Acknowledgements
We referred to [Dasheng](https://github.com/XiaoMi/Dasheng) and [Vocos](https://github.com/gemelo-ai/vocos) to implement this.

# Citation

```bibtex
@inproceedings{xingwei2025dashengdenoiser,
  title={Efficient Speech Enhancement via Embeddings from Pre-trained Generative Audioencoders},
  author={Xingwei Sun, Heinrich Dinkel, Yadong Niu, Linzhang Wang, Junbo Zhang, Jian Luan},
  booktitle={Interspeech 2025},
  year={2025}
}
```