In this work, we propose CleanMel, a single-channel Mel-spectrogram denoising and dereverberation network for improving both speech quality and automatic speech recognition (ASR) performance. The proposed network takes as input the noisy and reverberant microphone recording and predicts the corresponding clean Mel-spectrogram. The enhanced Mel-spectrogram can be either transformed to speech waveform with a neural vocoder or directly used for ASR. The proposed network is composed of interleaved cross-band and narrow-band processing in the Mel-frequency domain, for learning the full-band spectral pattern and the narrow-band properties of signals, respectively. Compared to linear-frequency domain or time-domain speech enhancement, the key advantage of Mel-spectrogram enhancement is that Mel-frequency presents speech in a more compact way and thus is easier to learn, which will benefit both speech quality and ASR. Experimental results on four English and one Chinese datasets demonstrate a significant improvement in both speech quality and ASR performance achieved by the proposed model.
Demos
Please open this page with Edge or Chrome. The audio playing is problematic in Firefox.
Mode
Method
CHiME
REVERB
DNS
EARS
RealMAN static
real
simu.
real
simu.
w.o. reverb.
w. reverb.
Unprocessed
Oracle
Online
FullSubNet
Demucs(dns64)
oSpatialNet
CleanMel-S-map (prop.)
CleanMel-S-mask (prop.)
Offline
FullSubNet
VoiceFixer
StoRM
SpatialNet
SpatialNet-Mamba
CleanMel-S-map (prop.)
CleanMel-S-mask (prop.)
CleanMel-L-mask (prop.)
Citation
@misc{shao2025cleanmel,
title={CleanMel: Mel-Spectrogram Enhancement for Improving Both Speech Quality and ASR},
author={Nian Shao and Rui Zhou and Pengyu Wang and Xian Li and Ying Fang and Yujie Yang and Xiaofei Li},
year={2025},
eprint={2502.20040},
archivePrefix={arXiv},
primaryClass={eess.AS},
url={https://arxiv.org/abs/2502.20040}
}
