Information and Communication College, Guilin University of Electronic and Technology, Guilin 541010, China
Show less
文章历史+
Received
Published
Revised
2025-03-17
2028-07-28
2025-11-04
Online Date
2026-09-20
PDF (8266K)
摘要
作为信号处理的重要研究方向,语音增强在人际交流、辅助破案和军事领域中扮演着至关重要的角色。针对现有深度学习方法未充分利用相位信息且特征提取单一导致增强语音质量下降的问题,提出了一种基于复数SepFormer与多尺度卷积的单通道语音增强方法(monaural speech enhancement based on complex SepFormer and multi-scale convolution,CS-MSC)。首先,引入了多尺度特征提取模块,改善传统方法特征提取单一的问题,并增强了模型对高频和低频特征的捕获能力,有效提升高频细节增强效果;其次,设计了带有通道注意力机制的跳跃连接,防止信息在深层网络中丢失,缓解深层网络中的梯度消失问题;并基于复数谱,在瓶颈层对振幅和相位的关联性进行建模,克服传统方法忽视相位信息的缺陷;最后,通过在编码器和解码器中添加增强频率轴特征表示的深度卷积提升模型在频率轴上的特征提取能力。实验结果表明,相比于门控卷积循环网络(gated convolution recurrent network,GCRN)、深度复数卷积循环网络(deep complex convolution recurrent network,DCCRN)等同类语音增强网络,本文提出的网络在极低信噪比条件下显著提高了语音信号的质量、可懂度以及信噪比;相较于DCCRN,在VoiceBank-Demand数据集上的语音感知质量和综合质量测度分别提升了31.72%和22.02%,表明该网络能够有效提升语音的可懂度和整体质量,具有较为突出的鲁棒性与泛化能力。
Abstract
As a critical research area in signal processing, speech enhancement plays a vital role in interpersonal communication, criminal investigation assistance, and military applications.To address limitations in existing deep learning approaches, including insufficient utilization of phase information and single-scale feature extraction leading to degraded speech quality, a monaural speech enhancement based on complex SepFormer and multi-scale convolution (CS-MSC) was proposed. To resolve the restricted feature diversity in conventional methods,a multi-scale feature extraction module was introduced to enhance the capture of both high- and low-frequency features and improve high-frequency detail reconstruction.Furthermore, skip connections with channel attention mechanisms were designed to mitigate information loss in deep networks and alleviate gradient vanishing. Additionally, complex-valued spectral representations were employed at the bottleneck layer to model amplitude-phase correlations, addressing the common neglect of phase information in conventional approaches.Finally, deepthwise convolution layers were incorporated into the encoder-decoder architecture to strengthen feature characterization along the frequency axis.Experimental results demonstrate that, compared to similar speech enhancement networks such as gated convolution recurrent network (GCRN) and deep complex convolution recurrent network (DCCRN), the proposed network significantly improves speech quality, intelligibility, and signal-to-noise ratio (SNR) under extremely low SNR conditions. Compared to DCCRN, the proposed network achieves improvements of 31.72% in perceptual evaluation of speech quality and 22.02% in composite quality measure on the VoiceBank-DEMAND dataset, effectively enhancing speech intelligibility and overall quality, and demonstrating notable robustness and generalization capabilities.
LI Zhou, LIU Qing-hua.
Monaural Speech Enhancement Method Based on Complex SepFormer and Multi-scale Convolution[J].
Science Technology and Engineering, 2026, 26(3): 1075-1086 DOI:10.12404/j.issn.1671-1815.2501786
时频掩蔽研究方面,Wang[7]首先提出了基于理想二值掩蔽(ideal binary mask,IBM)的语音增强方法,通过计算IBM将语音增强视为二分类任务。虽然IBM已经被证明可以显著提高语音清晰度,但是以IBM为目标的方法由于置零操作会导致语音失真,降低语音质量。候聪颖等[8]引入了以理想比率掩蔽(ideal ratio mask,IRM)和相位灵敏度掩蔽(phase sensitive mask,PSM)为目标的方法,其中IRM只作用于带噪语音的幅度谱,并直接使用带噪语音的相位谱来重建时域波形,忽略了相位信息的建模,而PSM虽联合优化相位,但计算复杂度较高。高键等[9]以复数理想比率掩蔽(complex ideal ratio mask,cIRM)为目标,在复数域中同时增强幅度谱和相位谱解决了相位问题,但训练时对神经网络模型要求较高。
为了解决上述问题,现提出一种基于复数SepFormer与多尺度卷积的语音增强方法(monaural speech enhancement based on complex SepFormer and multi-scale convolution,CS-MSC),引入一个编码器和两个解码器的结构对复数特征进行编解码,并使用多尺度卷积提取特征的多层次信息,在瓶颈层中对SepFormer内部进行复数乘法运算改进,充分挖掘复数特征中实部和虚部的关联性,提高模型对长序列信息的表达能力。研究主要工作如下。
为了全面评估CS-MSC模型的表现,在TIMIT语料库以及VoiceBank-Demand数据集(VBD)上进行实验,并与经典语音增强模型GCRN、DCCRN等进行对比。实验采用多种客观评价指标,包括语音感知质量(perceptual evaluation of speech quality,PESQ)、短时客观可懂度(short-time objective intelligibility,STOI)、总体信噪比(signal-to-noise ratio,SNR)、信号失真测度(mean opinion score prediction of the speech distortion,CSIG)、噪声失真测度(mean opinion score prediction of the intrusiveness of background noise,CBAK)以及综合质量测度(mean opinion score prediction of the overall processed speech quality,COVL),以衡量增强语音的质量、可懂度和噪声抑制效果。
MengXin, MaJianfen, ZhangXueying, et al. Speech enhancement algorithm based on the noise characteristics[J]. Science Technology and Engineering, 2016, 16(33): 244-248, 261.
HuJinyan, LiSheng, ChenFuming, et al. A new kind of nonlinear spectral subtraction method for radar speech enhancement[J]. Science Technology and Engineering, 2017, 17(16): 76-81.
[5]
TanK, WangD. Learning complex spectral mapping with gated convolutional recurrent networks for monaural speech enhancement[J]. IEEE/ACM Transactions on Audio, Speech, and Language Processing, 2020, 28: 380-390.
ZhangTianqi, LuoQingyu, ZhangHuizhi, et al. Speech enhancement method based on complex spectrum mapping with efficient transformer[J]. Journal of Signal Processing, 2024, 40(2): 406-416.
[8]
BorgstromB, BrandsteinM. Speech enhancement via attention masking network (SEAMNET): an end-to-end system for joint suppression of noise and reverberation[J]. IEEE/ACM Transactions on Audio, Speech, and Language Processing, 2021, 29: 515-526.
[9]
WangY, WangD. Towards scaling up classification-based speech separation[J]. IEEE Transactions on Audio, Speech, and Language Processing, 2013, 21(7): 1381-1390.
[10]
WangD. On ideal binary mask as the computational goal of auditory scene analysis[M] Boston: Springer, 2005: 181-197.
HouCongying, YangWenqing, WangZhao, et al. Speech enhancement based on time-frequency self-attention residual temporal convolutional networks[J]. Computer and Modernization, 2024(9): 20-24.
GaoJian, LiJunfeng. Complex convolutional recurrent network based speech enhancement[J]. Network New Media Technology, 2022, 11(1): 14-19, 42.
[15]
ZhaoH, ZararS, TashevI, et al. Convolutional-recurrent neural networks for speech enhancement[C]// Processings of the 2018 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). Calgary: IEEE, 2018: 2401-2405.
[16]
HochreiterS, SchmidhuberJ. Long short-term memory[J]. Neural Computation, 1997, 9(8): 1735-1780.
[17]
KounovskyT, MalekJ. Single channel speech enhancement using convolutional neural network[C]// Processings of the 2017 IEEE International Workshop of Electronics, Control, Measurement, Signals and their Application to Mechatronics (ECMSM). Donostia: IEEE, 2017: 1-5.
[18]
MikolovT, KarafiátM, BurgetL, et al. Recurrent neural network based language model[C]// Proceedings of the 2010 Interspeech. Chiba: IEEE, 2010: 1045-1048.
[19]
TanK, WangD. A convolutional recurrent neural network for real-time speech enhancement[C]// Proceedings of the 2018 Interspeech. Hyderabad: IEEE, 2018: 3229-3233.
[20]
HuY, LiuY, LüS, et al. DCCRN: deep complex convolution recurrent network for phase-aware speech enhancement[C]// Proceedings of the 2020 Interspeech. Shanghai: IEEE, 2020: 2472-2476.
[21]
ZhaoS, BinM, KarnW, et al. FRCRN: boosting feature representation using frequency recurrence for monaural speech enhancement[C]// Processings of the 2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). Singapore: IEEE, 2022: 9281-9285.
[22]
VaswaniA, ShazeerN, ParmarN, et al. Attention is all you need[J]. arXiv Preprint, arXiv: 1706. 03762, 2017.
FanJunyi, YangJibin, ZhangXiongwei, et al. Research on transformer-based single-channel speech enhancement[J]. Computer Engineering and Applications, 2022, 58(12): 25-36.
ZhuZhenfei, GeDongyuan, YaoXifan, et al. Environmental sound classification model combining Swin Transformer and CNN[J]. Science Technology and Engineering, 2024, 24(28): 12259-12267.
[27]
GulatiA, QinJ, ChiuC, et al. Conformer: convolution-augmented transformer for speech recognition[C]// Proceedings of the 2020 Interspeech. Shanghai: IEEE, 2020: 5036-5040.
CaoJie, WangChenzhang, LiangHaopeng, et al. Speech enhancement method based on two-branch attention and U-Net[J]. Application Research of Computers, 2024, 41(4): 1112-1116.
[30]
CemS, MircoR, SamueleC, et al. Attention is all you need in speech separation[C]// Processings of the 2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). Toronto: IEEE, 2020: 21-25.
[31]
YangM, MartinQ, LiD, et al. Complex Transformer: a framework for modeling complex-valued sequence[C]// Processings of the 2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). Barcelona: IEEE, 2019: 4232-4236.
[32]
PaliwalK, WojcickiK, ShannonB. The importance of phase in speech enhancement[J]. Speech Communication, 2011, 53(4): 465-494.
[33]
TanK, WangD. Complex spectral mapping with a convolutional recurrent network for monaural speech enhancement[C]// Processings of the 2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). Brighton: IEEE, 2019: 6865-6869.