基于特征流融合的带噪语音检测算法

doi:10.11959/j.issn.1000-436x.2020067

Abstract

Abstract:

Aiming at the problem that the initial detection performance of voice segment was poor,and the voice continuity structure was damaged in voice communication,a noisy voice detection algorithm based on feature stream fusion was proposed.Firstly,the time domain feature stream,the spectral pattern feature stream and the statistical feature stream were extracted according to the voice characteristics.Secondly,the voice segment in the noisy audio was estimated by different voice feature streams.Finally,the voice prediction probability obtained by each feature stream was weighted and fused,and the voice estimation probability was processed in short time by the hidden Markov model.The performance test of composite voice database under the condition of multi-type noise and different signal-to-noise ratio shows that compared with the baseline model based on Bayesian and DNN classifier,the voice detection accuracy of the proposed algorithm is improved by 21.26% and 11.01% respectively,and the quality of target voice is significantly improved.

Key words: voice communication, voice detection, feature stream fusion, hidden Markov model

CLC Number:

TP391.42

Hua LONG,Mingliang YANG,Yubin SHAO. Noisy voice detection algorithm based on feature stream fusion[J]. Journal on Communications, 2020, 41(4): 134-142.

Figures/Tables 10

References 22

[1]	WEI J , SUN X . Research on speech endpoint detection algorithm with low SNR[J]. Open Access Library Journal, 2017,4(3): 1-8.
[2]	LI Q , XIE H E , ZHENG Q J ,et al. The voice activity detection algorithm based on spectral entropy and high-order statistics[J]. Applied Mechanics and Materials, 2014(624): 495-499.
[3]	PALIWAL K K , WOJCICKI K , SCHWERIN B ,et al. Single-channe speech enhancement using spectral subtraction in the short-time modulation domain[J]. Speech Communication, 2010,52(5): 450-475.
[4]	MENDELEV V , PRISYACH T , PRUDNIKOV A ,et al. Robust voice activity detection with deep maxout neural networks[J]. Mathematical Models and Methods in Applied Sciences, 2015,9(8): 153-159.
[5]	LENGERICH C , HANNUN A . An end-to-end architecture for keyword spotting and voice activity detection[J]. arXiv Preprint,arXiv:1611.09405, 2016
[6]	IVRY A , BERDUGO B , COHEN I ,et al. Voice activity detection for transient noisy environment based on diffusion nets[J]. IEEE Journal of Selected Topics in Signal Processing, 2019,13(2): 254-264.
[7]	SOHN J , KIM N S , SUNG W ,et al. A statistical model-based voice activity detection[J]. IEEE Signal Processing Letters, 1999,6(1): 1-3.
[8]	FISHER E , TABRIKIAN J , DUBNOV S ,et al. Generalized likelihood ratio test for voiced-unvoiced decision in noisy speech using the harmonic model[J]. IEEE Transactions on Audio,Speech,and Language Processing, 2006,14(2): 502-510.
[9]	KIM J T , JUNG S H , CHO K H ,et al. Efficient harmonic peak detection of vowel sounds for enhanced voice activity detection[J]. Let Signal Processing, 2018,12(8): 975-982.
[10]	SHAMMA S A , ELHILALI M , MICHEYL C ,et al. Temporal coherence and attention in auditory scene analysis[J]. Trends in Neurosciences, 2011,34(3): 114-123.
[11]	HWANG S , JIN Y G , SHIN J W ,et al. Dual microphone voice activity detection based on reliable spatial cues[J]. Sensors, 2019,19(14): 3056-3064.
[12]	TEKI S , BARASCUD N , PICARD S ,et al. Neural correlates of auditory figure-ground segregation based on temporal coherence[J]. Cerebral Cortex, 2016,26(9): 3669-3680.
[13]	KLEINSCHMIDT M , . Spectro-temporal Gabor features as a front end for automatic speech recognition[C]// In 3rd European Congress on Acoustics.[S.n.:s.l]. 2002: 1-6.
[14]	罗智勇, 杨旭, 孙广路 ,等. 基于马尔可夫的有限自动机入侵容忍系统模型[J]. 通信学报, 2019,40(10): 79-89.
	LUO Z Y , YANG X , SUN G L ,et al. Finite automaton intrusion tolerance system model based on Markov[J]. Journal on Communications, 2019,40(10): 79-89.
[15]	NIELSEN J K , JENSEN T L , JENSEN J R ,et al. Fast fundamental frequency estimation:making a statistically efficient estimator computationally efficient[J]. Signal Process, 2017(135): 188-197.
[16]	SHE L M , NIELSEN J K , JENSEN J R ,et al. Bayesian pitch tracking based on the harmonic model[J]. IEEE/ACM Transactions on Audio,Speech,and Language Processing, 2019,27(11): 1737-1751.
[17]	SCHLUTER R , BEZRUKOV L , WANGNER H ,et al. Gammatone features and feature combination for large vocabulary speech recognition[C]// International Conference on Acoustics,Speech and Signal Processing. Piscataway:IEEE Press, 2007: 649-652.
[18]	SHAO Y , JIN Z , WANG D L ,et al. An auditory-based feature for robust speech recognition[C]// Proceedings of International Conference on Acoustics,Speech and Signal Processing. Piscataway:IEEE Press, 2009: 4625-4628.
[19]	MAARTEN V S , ANDREAS T , NARAYANAN S . A robust front end for VAD:exploiting contextual,discriminative and spectral cues of human voice[C]// Proceedings of the Annual Conference of the International Speech Communication Association. Piscataway:IEEE Press, 2013: 704-708.
[20]	MAARTEN V S , ANDREAS T ， NARAYANAN . A robust front end for VAD:exploiting contextual,discriminative and spectral cues of human voice[C]// Proceedings of the Annual Conference of the International Speech Communication Association. Piscataway:IEEE Press, 2013: 704-708.
[21]	LU L , JIANG H , ZHANG H ,et al. A robust audio classification and segmentation method[C]// ACM Multimedia. New York:ACM Press, 2001: 203-211.
[22]	ZHAO J H , GAO H B , LIU Y C ,et al. Speech recognition algorithm based on neural network and hidden Markov model[J]. The Journal of China Universities of Posts and Telecommunications, 2018,25(4): 28-37.

Metrics

Recommended 0

No Suggested Reading articles found!

数据集	语音数据/句			噪声数据/句
数据集	TIMIT	THCHS30	DRC	SBNL	Nonspeech
训练集	4 620	220	3 000	80	0
测试集	1 680	200	500	0	100

检测方法	SNR=-5 dB	SNR=0 dB	SNR=5 dB	SNR=10 dB
BFS	0.206 4	0.197 2	0.162 1	0.118 3
SCS	0.219 4	0.198 4	0.165 3	0.153 8
FSF	0.194 1	0.185 8	0.175 1	0.122 2

检测方法	SNR=-5 dB	SNR=0 dB	SNR=5 dB	SNR=10 dB
BFS	0.194 7	0.175 9	0.186 7	0.193 6
SCS	0.127 6	0.116 9	0.110 6	0.106 1
FSF	0.094 0	0.057 3	0.032 9	0.028 1

检测方法	SNR=-5 dB	SNR=0 dB	SNR=5 dB	SNR=10 dB
BFS	0.598 9	0.626 9	0.651 2	0.688 1
SCS	0.653 0	0.684 7	0.724 1	0.740 1
FSF	0.711 9	0.756 8	0.792 0	0.849 7

检测方法	Pf	Pm	Pc
BFS	0.171 050	0.187 725	0.641 275
SCS	0.184 225	0.115 300	0.700 475
FSF	0.169 300	0.053 0475	0.777 600

Noisy voice detection algorithm based on feature stream fusion

RichHTML

PDF

Knowledge

Abstract

Cite this article

share this article

Figures/Tables 10

References 22

Related Articles 9

Metrics

Recommended 0

[1]	Rong QIAN, Jianting XU, Kejun ZHANG, Hongyu DONG, Fangyuan XING. Research on HMM based link prediction method in heterogeneous network [J]. Journal on Communications, 2022, 43(5): 214-225.
[2]	Yuanbo GUO,Chunhui LIU,Jing KONG,Yifeng WANG. Study on user behavior profiling in insider threat detection [J]. Journal on Communications, 2018, 39(12): 141-150.
[3]	Jian-zhen LUO,Shun-zheng YU,Jun CAI. Method for determining the lengths of protocol keywords based on maximum likelihood probability [J]. Journal on Communications, 2016, 37(6): 119-128.
[4]	Chang-hai WANG,Jian-zhong ZHANG,Jing-dong XU,Yu-wei XU. Identifying the confidence level of activity recognition via HMM [J]. Journal on Communications, 2016, 37(5): 143-151.
[5]	Kai ZHONG,Hua PENG,Lin-dong GE. Blind Turbo equalization algorithm based on hidden Markov model for continuous phase modulation signals [J]. Journal on Communications, 2015, 36(3): 170-178.
[6]	Sheng-zong LIU,Xiao-ping FAN,Zhi-fang LIAO,Jia HU. Hidden Markov model fused with staying time for personalized recommendation [J]. Journal on Communications, 2014, 35(9): 112-121.
[7]	Xue-tao DUAN,Chun-fu JIA,Chun-bo LIU. Intrusion detection method based on hierarchical hidde kov model and variable-length semantic pattern [J]. Journal on Communications, 2010, 31(3): 109-114.
[8]	Chun-tao WANG,Jiang-qun NI,Ji-wu HUANG,Rong-yue ZHANG,Xi-zhang LUO. Robust DWT-HMM image watermarking algorithm optimized with content-adaptive approach [J]. Journal on Communications, 2007, 28(2): 80-87.
[9]	Wei-wei YIN,Le-nan WU. Joint source-channel iterative decoding based on hidden markov model [J]. Journal on Communications, 2006, 27(7): 61-65.