复杂环境下基于时延估计的声源定位技术研究

doi:10.3969/j.issn.1000-436x.2014.01.021

摘要/Abstract

摘要：

为了改善在复杂环境下声源定位算法的性能，提出了一种新的时延估计（TDE）方法，即基于传递函数比的统计模型方法（ATFR-SM）。该方法采用统计模型去除噪声对传递函数（ATF）的影响，在计算传递函数时对功率谱密度（PSD）进行平滑和“白化”，以去除混响对传递函数的影响。同时，算法中引入话音激活检测（VAD）去除对求取传递函数无用的噪声段，以提高时延估计的准确性。此外，将所提时延估计方法与线性定位法相结合，构成一套完整的声源定位方法。实验结果表明，在复杂环境下，时延估计方法具有更低的异常点百分比（PAP）和均方根误差（RMSE），且明显优于传统的参考算法，同时声源定位方法具有更高的定位精度。

关键词: 时延估计, 传递函数比, VAD, 统计模型, 声源定位

Abstract:

In order to improve the performance of source localization in noisy and reverberant environments,a novel time delay estimation (TDE) method was proposed.This method is called acoustical transfer function ratio based on statistical model (ATFR-SM).In the proposed algorithm,the noise reduction method based on the statistical model was adopted to reduce the effect of noise on acoustical transfer Function (ATF).In the ATF method,the power spectral density (PSD) was smoothed and whitened to reduce the effect of reverberations.voice activity detection (VAD) was used to distinguish the speech period from the noise period,and the TDE was performed in the speech period to improve the estimation accuracy.Moreover,the proposed TDE method and the linear closed-form method for source localization were combined to constitute a source localization system.The results of performance evaluation show that,in both the noisy and reverberant conditions,the lower percentage of abnormal points (PAP) and lower root mean square error (RMSE) can be achieved by the proposed TDE method than those of the reference methods.Meanwhile,the source localization has higher accuracy than the reference methods.

Key words: TDE, ATF ratio, VAD, statistical model, source localization

张大威,鲍长春,夏丙寅. 复杂环境下基于时延估计的声源定位技术研究[J]. 通信学报, 2014, 35(1): 183-190.

Da-wei ZHANG,Chang-chun BAO,Bing-yin XIA. Source localization based on time delay estimation in complex environment[J]. Journal on Communications, 2014, 35(1): 183-190.

图/表 11

图1

表1

图2

图3

图4

图5

图6

图7

表2

表3

表4

参考文献 17

[1]	SOUDEN M , BENESTY J , AFFES S . Broadband source localization from an eigenanalysis perspective[J]. IEEE Transactions on Audio,Speech,and Language Processing, 2010,18(6): 1575-1587.
[2]	GEDALYAHU K , ELDAR Y C . Time-delay estimation from low-rate samples:a union of subspaces approach[J]. IEEE Transactions on Signal Processing, 2010,58(6): 3017-3031.
[3]	SO H C , CHAN Y T , CHAN F K W . Closed-form formulae for time-difference-of-arrival estimation[J]. IEEE Transactions on Signal Processing, 2008,56(6): 2614-2620.
[4]	LUI K , CHAN F , SO H C . Semidefinite programming approach for range-difference based source localization[J]. IEEE Transactions on Signal Processing, 2009,57(4): 1630-1633.
[5]	KNAPP C , CARTER G . The generalized correlation method for estimation of time delay[J]. IEEE Transactions on Acoustics,Speech and Signal Processing, 1976,24(4): 320-327.
[6]	CHAMPAGNE B,BéDARD S,STéPHENNE A . Performance of time-delay estimation in the presence of room reverberation[J]. IEEE Transactions on Speech and Audio Processing, 1996,4(2): 148-152.
[7]	DVORKIND T G , GANNOT S . Approaches for time difference of arrival estimation in a noisy and reverberant environment[A]. Proceedings of the International Workshop on Acoustic Echo and Noise Control (IWAENC'03)[C]. Kyoto,Japan, 2003. 215-218.
[8]	CORNELIS B , DOCLO S , VAN DAN BOGAERT T ,et al. Theoretical analysis of binaural multimicrophone noise reduction techniques[J]. IEEE Transactions on Audio,Speech,and Language Processing, 2010,18(2): 342-355.
[9]	SALVATI D , CANAZZA S . Adaptive time delay estimation using filter length constraints for source localization in reverberant acoustic environments[J]. Signal Processing Letters, 2013,20(5): 507-510.
[10]	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.
[11]	COHEN I , BERDUGO B . Noise estimation by minima controlled recursive averaging for robust speech enhancement[J]. Signal Processing Letters,IEEE, 2002,9(1): 12-15.
[12]	LOIZOU P C . Speech enhancement based on perceptually motivated Bayesian estimators of the magnitude spectrum[J]. IEEE Transactions on Speech and Audio Processing, 2005,13(5): 857-869.
[13]	SOHN J , KIM N S , SUNG W . A statistical model-based voice activity detection[J]. Signal Processing Letters,IEEE, 1999,6(1): 1-3.
[14]	李光源, 崔慧娟, 唐昆 . 一种基于噪声估计的话音激活检测算法[J]. 信息技术, 2011 (10): 5-8. LI G Y , CUI H J , TANG K . An algorithm of voice activity detection based on noise estimation[J]. Information Technology, 2011,(10): 5-8.
[15]	ALLEN J B , BERKLEY D A . Image method for efficiently simulating small-room acoustics[J].The Journal of the Acoustical Society of America,1979,65:943. The Journal of the Acoustical Society of America, 1979,65:943.
[16]	VARGA A , STEENEKEN H J M.Assessment for automatic speech recognition:II . NOISEX-92:a database and an experiment to study the effect of additive noise on speech recognition systems[J]. Speech Communication, 1993,12(3): 247-251.
[17]	GILLETTE M D , SILVERMAN H F . A linear closed-form algorithm for source localization from time-differences of arrival[J]. Signal Processing Letters,IEEE, 2008,15: 1-4.

信噪比/dB	准确率/%	MDR/%	FAR/%
20	96.05	3.95	5.15
15	94.90	5.10	7.11
10	93.66	6.34	8.15
5	91.28	8.72	8.58
0	84.57	15.43	8.70
?5	71.15	28.85	9.74

混响时间	方法名称	估计坐标与真实坐标的绝对误差			估计坐标与真实坐标的标准差
混响时间	方法名称	x	y	z	σ_x	σ_y	σ_z
	参考方法1	0.043	0.035	0.031	0.231	0.216	0.172
200 ms	参考方法2	0.005	0.002	0.004	0.046	0.043	0.033
	提出方法	0.002	0.001	0.002	0.030	0.030	0.020
	参考方法1	0.065	0.052	0.047	0.472	0.356	0.241
400 ms	参考方法2	0.005	0.004	0.003	0.049	0.045	0.035
	提出方法	0.002	0.002	0.002	0.032	0.030	0.020

SNR	方法名称	估计坐标与真实坐标的绝对误差			坐标的标准差估计坐标与真实
SNR	方法名称	x	y	z	σ_x	σ_y	σ_z
	参考方法1	0.051	0.032	0.046	0.325	0.531	0.743
10 dB	参考方法2	0.005	0.003	0.004	0.045	0.065	0.078
	提出方法	0.001	0.001	0.001	0.019	0.012	0.023
	参考方法1	0.531	0.424	0.421	1.235	1.873	1.432
0 dB	参考方法2	0.042	0.033	0.031	0.426	0.923	0.565
	提出方法	0.021	0.012	0.011	0.234	0.476	0.254

测试环境	方法名称	估计坐标与真实坐标的绝对误差			估计坐标与真实坐标的标准差
测试环境	方法名称	x	y	z	σ_x	σ_y	σ_z
SNR=10 dBRT=200 ms	参考方法1	0.123	0.056	0.065	0.756	0.519	0.352
	参考方法2	0.022	0.003	0.005	0.054	0.043	0.025
	提出方法	0.005	0.001	0.002	0.023	0.020	0.014
SNR=0 dBRT=200 ms	参考方法1	0.212	0.124	0.235	0.785	2.318	0.893
	参考方法2	0.045	0.032	0.040	0.513	1.021	0.432
	提出方法	0.013	0.016	0.023	0.232	0.544	0.246
SNR=10 dBRT=400 ms	参考方法1	0.122	0.121	0.054	0.312	0.213	0.175
	参考方法2	0.021	0.020	0.008	0.057	0.051	0.046
	提出方法	0.002	0.004	0.003	0.021	0.030	0.011
SNR=0 dBRT=400 ms	参考方法1	0.322	0.265	0.098	0.878	1.765	1.543
	参考方法2	0.042	0.045	0.032	0.633	1.189	1.026
	提出方法	0.021	0.012	0.013	0.221	0.544	0.863