Model reference · open weights
dasheng-denoiser is an open-weight audio or speech model from mispeech. dasheng-denoiser (FP32) weighs 238 MB; the smallest configuration that runs it is RTX 3060 12 GB.
For self-hosting, compare this build's requirements below with the GPU you plan to rent.
Summary of the mispeech/dasheng-denoiser model card, 2026-10-01
What it is
| Released by | mispeech |
|---|---|
| Released | 2026-03-03 |
| Parameters | 119M |
| VRAM | 238 MB for the weights |
What it runs on
| Card | One stream | Memory |
|---|---|---|
| RTX 3060 12 GB | fits | 11.6 GB |
| RTX 4060 Ti 16 GB | fits | 15.4 GB |
| RTX 3090 24 GB | fits | 23.4 GB |
| RTX 4090 24 GB | fits | 23.4 GB |
| RTX 5090 32 GB | fits | 31.0 GB |
| L40S 48 GB | fits | 44.0 GB |
| A100 80 GB | fits | 78.2 GB |
| H100 80 GB | fits | 78.1 GB |
| RTX PRO 6000 Blackwell 96 GB | fits | 93.8 GB |
| DGX Spark (GB10) 128 GB unified | fits | 107 GB |
| H200 141 GB | fits | 138 GB |
| B200 180 GB | fits | 176 GB |
From the model card
Dasheng Denoiser Official PyTorch inference code for the Interspeech 2025 paper:
uv pip install transformers torch torchaudio einops
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)
We referred to Dasheng and Vocos to implement this.
@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}
}
Quoted from the model card on Hugging Face. The full card is behind the Hugging Face link above.