基于对数处理机制和时频掩蔽估计的语音增强

doi:10.16180/j.cnki.issn1007-7820.2025.01.007

电子科技 ›› 2025, Vol. 38 ›› Issue (1): 45-51.doi: 10.16180/j.cnki.issn1007-7820.2025.01.007

基于对数处理机制和时频掩蔽估计的语音增强

王显云(), 窦姗姗, 程楚皓

中国电子科技集团公司第三研究所声学部,北京 100015

收稿日期:2023-05-11 修回日期:2023-06-25 出版日期:2025-01-15 发布日期:2025-01-06
通讯作者: 王显云(1984-),男,E-mail:wang2005ji@126.com,博士,高级工程师。研究方向:语音与音频信号处理。
基金资助:
近地面探测技术重点实验室基金(6142414210405)

Speech Enhancement Based on the Logarithmic Processing and Time Frequency Masking Estimation

WANG Xianyun(), DOU Shanshan, CHENG Chuhao

Acoustic Department,China Electronics Technology Group Corporation Third Research Institute,Beijing 100015,China

Received:2023-05-11 Revised:2023-06-25 Online:2025-01-15 Published:2025-01-06
Supported by:
Science and Technology on Near-Surface Detection Laboratory(6142414210405)

摘要/Abstract

摘要：

针对时频谱模型估计语音不准确的问题,文中提出采用模型变换的方式来获得噪声和语音的对数概率密度函数,同时借助带噪语音、干净语音和噪声之间的对数关系并结合最小均方误差(Minimum Mean Square Error, MMSE)估计理论推导出估计语音对数谱的时频掩模。基于语音和噪声的对数概率分布推导出了一种软掩模,该软掩模可对带噪语音的对数子带进行加权以降低噪声,提高语音估计的准确性。仿真结果表明,与未处理的含噪语音相比,所提方法在噪声抑制方面具有3 dB以上的提升,基于最小均方误差的时频掩模和软掩模在听觉感知方面的平均提升量分别为27.7%和29.4%,在可懂度方面的平均提升量分别为12.7%和14.3%。

关键词: 语音处理, 语音增强, 对数概率密度函数, 时频掩模, 噪声抑制, 非监督学习, 软掩蔽, 对数谱

Abstract:

In view of the problem of inaccurate speech estimation by time-spectrum model, this study proposes a model transformation method to obtain the logarithmic probability density function of noise and speech. With the help of the logarithmic relationship among noisy speech, clean speech and noise, and the MMSE(Minimum Mean Square Error) estimation theory, a time-frequency mask of the estimating log-spectrum of speech is derived. A soft mask is also derived based on the logarithmic probability distribution of speech and noise, which can weight the logarithmic subbands of noisy speech to reduce noise and improve the accuracy of speech estimation. The simulation results show that compared with unprocessed noisy speech, the proposed method has an improvement of more than 3 dB in noise suppression. The average improvement in auditory perception of time-frequency mask and soft mask based on MMSE is 27.7% and 29.4%, and the average improvement in intelligibility is 12.7% and 14.3%, respectively.

Key words: speech processing, speech enhancement, logarithmic probability density function, time-frequency mask, noise suppression, unsupervised learning, soft mask, logarithmic spectrum

中图分类号:

TN912.35

王显云, 窦姗姗, 程楚皓. 基于对数处理机制和时频掩蔽估计的语音增强[J]. 电子科技, 2025, 38(1): 45-51.

WANG Xianyun, DOU Shanshan, CHENG Chuhao. Speech Enhancement Based on the Logarithmic Processing and Time Frequency Masking Estimation[J]. Electronic Science and Technology, 2025, 38(1): 45-51.

图/表 3

图1

表1

表2

参考文献 23

[1]	Ephraim Y, Malah D. Speech enhancement using a minimum mean-square error short time spectral amplitude estimator[J]. IEEE Transactions on Acoustics Speech and Signal Processing, 1984, 32(6):1109-1121.
[2]	Ephraim Y, Malah D. Speech enhancement using a minimum mean-square error log-spectral amplitude estimator[J]. IEEE Transactions on Acoustics Speech and Signal Processing, 1985, 23(2):443-445.
[3]	Loizou P. Speech enhancement based on perceptually motivated Bayesian estimators of the speech magnitude spectrum[J]. IEEE Transactions on Speech and Audio Processing, 2005, 13(2):857-869.
[4]	Wang D, Chen J. Supervised speech separation based on deep learning:An overview[J]. IEEE/ACM Transactions on Audio,Speech and Language Processing, 2018, 26(10):1702-1726.
[5]	Tan K, Wang D. 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(10):380-390.
[6]	Pascual S, Bonafonte A, Serrà J. SEGAN:Speech enhancement generative adversarial network[C]. Stockholm: Proceedings of the International Speech Communication Association, 2017:3642-3646.
[7]	李辉, 景浩, 严康华, 等. 基于卷积循环网络与非局部模块的语音增强方法[J]. 电子科技, 2022, 35(3):8-15.
	Li Hui, Jing Hao, Yan Kanghua, et al. Speech enhancement method based on convolutional recurrent network and non-local module[J]. Electronic Science and Technology, 2022, 35(3):8-15.
[8]	Roweis S T. One microphone source separation[C]. Boston: International Conference on Neural Information Processing Systems, 2001:793-799.
[9]	Jiu H, Hagai A. Speech enhancement,gain and noise spectrum adaptation using approximate Bayesian estimation[J]. IEEE Transactions on Acoustics Speech and Signal Processing, 2009, 17(1):24-37.
[10]	Xu Y, Dai L, Lee C. A regression approach to speech enhancement based on deep neural networks[J]. IEEE/ACM Transactions on Audio,Speech and Language Processing, 2015, 23(1):7-19.
[11]	Niu S, Du J, Chai L, et al. A maximum likelihood approach to multi-objective learning using generalized gaussian distributions for DNN-based speech enhancement[C]. Barcelona: IEEE International Conference on Acoustics,Speech and Signal Processing, 2020:6229-6233.
[12]	Chai L, Du J, Liu Qin F, et al. Using generalized gaussian distributions to improve regression error modeling for deep earning-based speech enhancement[J]. IEEE/ACM Transactions on Audio,Speech and Language Processing, 2019, 27(12):1919-1931.
[13]	何少芳, 周丽. 可搜索语音加密机制访问控制的匿名交互算法[J]. 电子科技, 2021, 34(4):1-5.
	He Shaofang, Zhou Li. Anonymous interaction algorithm of access control in searchable voice encryption mechanism[J]. Electronic Science and Technology, 2021, 34(4):1-5.
[14]	Papoulis A, Pillai S. Probability,random variables and stochastic processes[M]. New York: McGraw-Hill, 2002:319-326.
[15]	Radfar M, Banihashemi A. A non-linear minimum mean square error estimator for the mixture-maximization approximation[J]. Electron Letter, 2006, 42(12):75-76.
[16]	Nadas A, Nahamoo D, Picheny M. Speech recognition using noise-adaptive prototype[J]. IEEE Transactions on Acoustics Speech and Signal Processing, 1989, 37(10):1495-1505.
[17]	Hu K, Wang D. An iterative model-based approach to cochannel speech separation[J]. EURASIP Journal on Audio,Speech,Music Processing, 2013, 14(4):1-11.
[18]	Gradshteyn I, Ryzhik Z. Table of Integrals,Series and Products[M]. New York: Academic, 1980:174-193.
[19]	Martin R. Noise power spectral density estimation based on optimal smoothing and minimum statistics[J]. IEEE Transactions on Speech and Audio Processing, 2001, 9(5):504-512.
[20]	Mai V, Pastor D, Aissa A. Optimal Bayesian speech enhancement by parametric joint detection and estimation[J]. IEEE Access, 2020, 8(10):15695-15710.
[21]	Deng F, Bao F, Bao C. Speech enhancement using generalized weighted β-order spectral amplitude estimator[J]. Speech Communication, 2014, 59(1):55-68.
[22]	Rix A, Beerends J, Hollier M, et al. Perceptual evaluation of speech quality-A new method for speech quality assessment of telephone networks and codecs[C]. Piscataway: IEEE International Conference on Acoustics,Speech and Signal Processing, 2001:749-752.
[23]	Jensen J, Taal C. An algorithm for predicting the intelligibility of speech masked by modulated noise maskers[J]. IEEE Transactions on Audio Speech and Language Processing, 2016, 24(11):2009-2022.

噪声类型	信噪比/dB	不同方法
噪声类型	信噪比/dB	含噪	Ref.1	Ref.2	Ref.3	Ref.4	MASK_MMSE	MASK_P
White	-5	1.330	1.660	1.756	1.782	1.767	1.819	1.956
	0	1.564	2.058	2.185	2.217	2.189	2.265	2.384
	5	1.875	2.450	2.581	2.614	2.607	2.657	2.742
	10	2.249	2.813	2.937	2.965	2.968	2.993	3.064
Babble	-5	1.398	1.635	1.656	1.698	1.649	1.656	1.652
	0	1.791	2.025	2.060	2.082	2.049	2.070	2.069
	5	2.157	2.390	2.431	2.441	2.435	2.439	2.435
	10	2.502	2.724	2.764	2.771	2.766	2.772	2.768
F16	-5	1.379	1.823	1.918	1.940	1.911	1.962	2.021
	0	1.717	2.217	2.313	2.336	2.309	2.366	2.430
	5	2.092	2.577	2.681	2.707	2.684	2.738	2.801
	10	2.444	2.923	3.037	3.067	3.044	3.095	3.154
Factory	-5	1.345	1.718	1.787	1.780	1.775	1.808	1.820
	0	1.707	2.106	2.188	2.208	2.186	2.226	2.235
	5	2.092	2.501	2.584	2.604	2.602	2.615	2.614
	10	2.477	2.864	2.944	2.962	2.960	2.964	2.963
均值	-	1.882	2.281	2.364	2.385	2.368	2.403	2.435

噪声类型	信噪比/dB	不同方法
噪声类型	信噪比/dB	含噪/%	Ref.1/%	Ref.2/%	Ref.3/%	Ref.4/%	MASK_MMSE/%	MASK_P/%
White	-5	31.15	37.56	37.91	37.97	37.90	37.48	38.16
	0	43.77	50.42	51.00	51.11	50.95	50.91	52.04
	5	56.17	62.60	63.24	63.36	63.24	63.36	64.33
	10	68.01	73.78	74.32	74.41	74.42	74.43	74.69
Babble	-5	31.53	34.52	34.80	34.83	34.79	34.41	34.54
	0	45.78	49.66	50.22	50.33	50.20	50.04	50.39
	5	59.58	63.47	64.21	64.38	64.08	64.34	64.93
	10	71.34	74.57	75.19	75.35	75.17	75.30	75.80
F16	-5	30.60	37.63	38.39	38.51	38.18	37.98	39.37
	0	44.13	52.15	53.30	53.55	53.10	53.37	55.30
	5	57.77	65.78	67.09	67.41	66.91	67.60	69.47
	10	70.17	77.107	78.21	78.48	78.18	78.62	79.77
Factory	-5	32.29	38.36	38.90	38.91	38.74	38.45	38.92
	0	46.53	53.21	54.05	54.21	53.92	53.93	54.65
	5	60.49	66.59	67.58	67.80	67.34	67.80	68.56
	10	72.49	77.19	78.01	78.21	77.88	78.30	78.80
均值	-	51.36	57.16	57.89	58.05	57.81	57.89	58.72

基于对数处理机制和时频掩蔽估计的语音增强

Speech Enhancement Based on the Logarithmic Processing and Time Frequency Masking Estimation

RichHTML

PDF (PC)

赞

可视化

摘要/Abstract

引用本文

使用本文

图/表 3

参考文献 23

相关文章 4

Metrics

本文评价

推荐阅读 0

[1]	白兴宇,苟宇涛,姜煜,刘明禹. 一种复杂噪声环境下的机电系统故障在线监测声学处理方法[J]. 电子科技, 2023, 36(3): 55-61.
[2]	李辉,景浩,严康华,徐良浩. 基于卷积循环网络与非局部模块的语音增强方法[J]. 电子科技, 2022, 35(3): 8-15.
[3]	白兴宇,华生辉,姜煜,张敏. 一种基于宽带谱处理的机械系统故障检测方法[J]. 电子科技, 2021, 34(12): 25-29.
[4]	何佶珂, 曾晓东. 光外差平衡检测与本振光强度噪声抑制研究[J]. , 2010, 23(8): 25-.