Speech enhancement plays an important role in numerous speech applications with purpose of preserving the clean speech data while suppressing the background noise, interference and third-party talker. Singlechannel approach, which bases on the spectral subtraction, owns the ability of noise reduction in stationary noise environments. Unfortunately, due to the rapidly changing surrounding noise, the existence of nondirectional noise, non-stationary noise degrades the performance of signal processing system and causes the speech distortion. Therefore, microphone array (MA) technology has been implemented into various types of acoustic equipment for extracting the target desired speech component while alleviating background noise. MA exploits the spatial priori information of designed configuration of MA, the characteristics of environment, the properties of processed captured MA signals to obtain a high directional beampattern towards at the specified direction. Generalized sidelobe canceller (GSC) beamformer is a very effective beamforming technique for both speech enhancement and noise reduction simultaneously. However, because of the difference microphone sensitivities, the error of sampling rate, the microphone output signal. In this paper, the author proposed an effective post-Filtering for mitigating noise components and increasing the signal-to-noise ratio. The demonstrated experiment has confirmed the capability of the suggested method in realistic recording scenarios to enhance the signal-to-noise ratio from 12.7 to 15.0 dB. -Filtering can be integrated into multi-channel system for addressing other complicated problems, such as, automatic speech recognition, reverberation.
Because of the influence of third-party talker, the complex and annoying environment, the existence of various types of noise, the observed speech signals often corrupted and always polluted by noise, which leads to the speech quality, speech intelligibility deteriorated, and perceptual metric listener are significantly affected, as in Figure 1. Therefore, the requirement of reducing noise and improving engineering to study and illustrate experiment.
Single - channel approach is an efficient technique to extract the original speech component in stationary noise. However, this direction often based on spectral subtraction, which does not correctly outperform in non - stationary and complex noise field, cause speech distortion and degrades the satisfactory of listener. In recent years, MA technology has been popular installed in numerous speech applications, such as, surveillance device, teleconference system, smart - home, voice - controlled equipment, cochlear implant, hearing aids.
MA beamforming uses the prior spatial information, the designed configuration of microphone distribution to obtain the steerable beampattern towards the sound source while suppressing the background noise, interference or third - party speaker from other directions. MA beamforming includes different techniques: Delay - and - sum (DAS) [1], Differential microphone array (DIF) [24], Minimum variance distortionless response (MVDR) [5-7], Linearly constraint minimum variance (LCMV) [8-10] and GSC beamformer [10-18]. These beamforming methods exploit the preferred steering vector or optimum constrained criteria of minimizing the output noise power while saving the clean speech data, as in Figure 2.
GSC beamformer concerns the beampattern toward the specified direction and uses an adaptive noise canceller (ANC) for saving the speech component. GSC owns high directivity index, high performance in various types of noise fields. However, in realistic recording scenario, due to the efforts are studied to address this problem. The principal working of MA beamforming can be given in Figure 3.
Li S [11] proposed usin to control the step size for adjusting the weight coefficients of adaptive noise suppression. The numerical simulation has confirmed the effectiveness of the suggested technique in diffuse noise field with improving the speech quality.
Wang J [12] described the adaptation control of ANC, which follows a time - varying Gaussian distribution. The advantage of this approach is calculating the target speech variance and coefficient according to the maximum likelihood criterion. The conducted experiments were demonstrated under various conditions to verify the robustness and effectiveness of the proposed method.
In [13], the external microphones were implemented to GSC structure to minimum the power distortionless response beamformer and suppress interferer. The numerical simulations shown the better speech enhancement in adverse recording scenario.
Dai S [14] proposed an efficient algorithm for generating blocking matrix of GSC beamforming. The approach applies simplified zero placement algorithm to obtain independent full space vectors. In comparison with conventional singular value decomposition, the suggested method owns more constraints in numerous situations.
Park J [15] proposed method with two stages in GSC beamformer, which contains determinant technique employs signal activity detector to increase the overall performance. The demonstrated experiment has confirmed the effectiveness through spectrogram, mean opinion score and objective measures.
Li B [16] suggested new optimization algorithm for removing leakage of speech caused by inaccurate estimation of direction of arrival (DoA) and the correlation between the output signal of simulation results show that the proposed technique has better noise suppression and speech enhancement.
In [17], the feature of interaction between adaptive beamforming and multi-channel post-filtering applied for achieving more robust adaptive beamforming in adverse environment. The experiment show that suggested direction obtains considered speech enhancement, improvement in the terms of noise suppression.
Li J [18] incorporates frequency domain independent component analysis and GSC beamformer to perform a steep null in desired target talker. The experimental and numerical simulation show the adverse environment with modifying the internal structure, the control of blocking matrix or the ANC rule. However, because of the microphone mismatches, the different microphone sensitivities, the error of sampling frequency, the displacement of MA, the moving head of speaker, which distort speech component. The above research cannot properly outperform in complex and annoying conditions. In this paper, the author proposed a modified Wiener filter, which works well in realistic recording situation for suppressing the musical noise, residual noise, considerable noise level and improving the signal-toapplied into real-time impact acoustic equipment, which does not require larger database or multipproach is introducing an easy executed technique, which can adverse or complex environment. This method does not require large database for training or testing.
The rest of this paper is organized as: Section II presents the GSC beamformer structure and the
Filtering, Section IV illustrates the experiment in realistic experiment, Section V concludes and the
2. Generalized sidelobe canceller beamformer
As in Figure 4, in this section, the author describes the representation of GSC beamformer by using
dual-microphone system (DMA2). In the short-time Fourier transform, the formulation of received
array signals 1( , ), 2( , ) can be presented at current frequency , current frame , as:
1( , ) = ( , ) + 1( , )
2( , ) = ( , ) − + 2( , )
(
GSC structure contains three parts: the fixed beamformer, and blocking matrix and ANC. The fixed beamformer steer the beampattern at specified direction of target speech source, while blocking matrix attenuates the speech component for obtaining the only noise, and ANC extracts the desired AS beamformer, due to its simplicity computation. DIF is applied for blocking matrix to achieve the only noise component and Wiener filter is implemented as adaptive signal processing algorithm.
The output of fixed beamformer and blocking matrix is ( , ), ( , ) respectively. The
formulation of ( , ), ( , ) can be expressed as:
( , ) =
1( , ) − + 2( , )
2
(
The Wiener filter formulation: following equations:
In realistic recording scenario, due to complex and annoying recording scenario, microphone mismatches, the error of estimation of steering vector, the displacement of microphone geometry, -life situations constantly degraded. Therefore, the following section describes the suggested approach for further suppressing remained noise at GSC where * is the asterisk operator.
The auto-cross power spectral densities of ( , ), ( , ) and ( , ) can be calculated as the
( , ) = ( , ) = ( , − 1) + (1 − ) ( , ) ∗( , ) ( , − 1) + (1 − ) ( , ) ∗( , ) where is the smoothing parameter in the range {0 … 1}.
The final output signal can be derived as the following equation: ( , ) = ( , ) − ( , )
( , ) ( , ) = 1( , ) − − 2( , ) ( , ) = { ( , ) ∗( , )} { ( , ) ∗( , )} ( , ) is applied for extracting the desired target speaker as the ( , ) = 2( , ) = ( , ) −1 ( , ) ( , )
1 ( , ) = 0.5 + 0.5
12 (2 )
( , ) = 1 −
12 (2 )
(
The Wiener filter - based post-Filtering has the formulation as:
- based traditional post Filtering according to the Where 2( , ) is variance of speech and 2( , ) is the variance of noise.
By applying the microphone array beamforming, the variance of speech can be computed as: where ( , ) = [ − ] ,
( , ) = { ( , ) ( , )}, ( , ) is the covariance matrix of observed microphone array signals, ( , ) = [ 1( , ) 2( , )] , and 2( , ) is variance of noise, which can be calculated by minimum statistic [19].
Based on the prior information of the impinging incident of speaker, , the author uses the spectral gain ( , ) and
( , ) [20] which determined as: Where (2 ) = 12. . , 12 is the phase difference of two received array signals.
- Filtering is combining the Wiener filter - based traditional postFiltering and the spectral gain to form an appropriate post-Filtering as: ( , ) = where is a smoothing parameter, in the range of {0 … 1}.
The final output signal of GSC beamformer is filtered as: ̂ ( , ) = ( , ) × ( , ) (13) (14) arrival of helpful signal, the phase difference to adaptively control the Wiener filter - based post Filtering for suppress residual noise, a certain noise level. The effectiveness of this approach can be depicted in the next section.
The purpose of this section is illustrating the experiment in real - life recording scenario to demonstrate the efficient post-Filtering in reducing residual noise, surrounding noise and enhancing the overall speech enhancement of GSC beamformer. The author uses an objective measurement [20] for calculating the SNR between the observed MA signals, the processed signals by traditional is given in Figure 5, in living room 5 4.5 3.5 ( ) with existence of third - party talker, washing machine, non-directional noise.
A speaker stands at distance = 5( ) to dual - microphone system (DMA2), the impinging incident angle of helpful signal is = 90 (
) relatives to the axis of DMA2, the range between two mounted microphones is = 5(
). For capturing the original speech data, sampling frequency , overlap 50%. The received array signals can be shown in Figure 6 and Figure 7. -Filtering (appF). The scheme of experiment
With smoothing parameter = 0.1, the output signal of tGB is derived as:
Due to the complex and adverse environment, the undetermined acoustical factors, the different microphone sensitivities, the error of sampling frequency, the moving head of talker, microphone residual noise, speech distortion is challenging task in almost speech applications.
Therefore, efficient post-Filtering allows removing this drawback and enhancing the speech quality. The promising result by implementing appF with = 0.98 is given in Figure 10 and Figure 11. ng the spectral gain of MA beamforming to modified the post-Filtering, which based on Wiener filter. Because of the complex condition, the presence and absence speech component cause the difficulty of computing Wiener filter, this approach compensates the distortion of Wiener filter. The promising result of the above method is improving the accuracy of Wiener filter according to the presence/absence of speech component. The numerical simulation shows noise reduction and preserving clean speech data. The described technique can be integrated to dual-microphone system for immediately addressing speech enhancement problem in real-life recording environment.
WADA (Waveform Amplitude Distribution Analysis) SNR [21] based on the constrained criteria of the speech component distributed according to Gamma model and additive Gaussian noise. Based on sequential Gaussian mixture estimation, NIST (National Institute of Standards and Technology) [22] calculates the STNR (Signal-To-Noise Ratio).
Microphone technology has been commonly used in numerous acoustic equipment, such as teleconference system, hearing aids, voice - controlled device, smart-phone, cochlear implant, surveillance device. MA beamforming uses the priori information of spatial distribution, designed configuration to obtain the advantage of noise reduction, speech enhancement simultaneously with acceptable perceptual metric listener, speech quality. GSC beamformer owns the high directivity index at the specified direction and exploits an adaptive noise canceller to extract the desired target speaker while suppressing background noise. In this paper, the author proposed using an efficient postachieve the clean speech data with satisfied speech intelligibility. The approach uses the prior information of preferred steering vector, the spectral gain to modify essential signal-to-noise ratio to form an accurate post-Filtering. The numerical simulations have confirmed the effectiveness of evel to 12.5 dB and increasing the speech quality from 12.7 to 15.0 dB. The promising result has verified the capability of the suggested postFiltering under realistic recording condition and this approach can be integrated into multi-channel signal processing system for dealing other complicated problems, such as speech recognition, reverberation. In the future, the author will incorporate the characteristic of environment to enhance post-Filtering.
The author(s) have not employed any Generative AI tools. Data Science and Information Technology (DSIT), Shanghai, China, 2022, pp. 1-6, doi: 10.1109/DSIT55514.2022.9943855. [12] J. Wang, F. Yang, J. Guo and J. Yang, "Robust Adaptation Control for Generalized Sidelobe Canceller with Time-Varying Gaussian Source Model," 2023 31st European Signal Processing Conference (EUSIPCO), Helsinki, Finland, 2023, pp. 16-20, doi: 10.23919/EUSIPCO58844.2023.10289801. [13] W. Middelberg and S. Doclo, "Comparison of Generalized Sidelobe Canceller Structures Incorporating External Microphones for Joint Noise and Interferer Reduction," Speech Communication; 14th ITG Conference, online, 2021, pp. 1-5. [14] S. Dai, M. Li, Q. H. Abbasi and M. Imran, "A Fast Blocking Matrix Generating Algorithm for Generalized Sidelobe Canceller Beamforming," 2020 IEEE International Symposium on Antennas and Propagation and North American Radio Science Meeting, Montreal, QC, Canada, 2020, pp. 291-292, doi: 10.1109/IEEECONF35879.2020.9329735. [15] J. Park, J. Hong, J. -W. Choi and M. Hahn, "Determinant-Based Generalized Sidelobe Canceller for Dual-Sensor Noise Reduction," in IEEE Sensors Journal, vol. 22, no. 9, pp. 8858-8868, 1 May1, 2022, doi: 10.1109/JSEN.2022.3162619. [16] B. Li and L. -H. Zhang, "An improved speech enhancement algorithm based on generalized sidelobe canceller," 2016 International Conference on Audio, Language and Image Processing (ICALIP), Shanghai, China, 2016, pp. 463-468, doi: 10.1109/ICALIP.2016.7846528. [17] K. Li, Q. Fu and Y. Yan, "Speech Enhancement Using Robust Generalized Sidelobe Canceller with Multi-Channel Post-Filtering in Adverse Environments*," in Chinese Journal of Electronics, vol. 21, no. 1, pp. 85-90, January 2012. [18] J. Li, Q. Fu and Y. Yan, "An approach of adaptive blocking matrix based on frequency domain independent component analysis in generalized sidelobe canceller," IEEE 10th INTERNATIONAL CONFERENCE ON SIGNAL PROCESSING PROCEEDINGS, Beijing, China, 2010, pp. 231-234, doi: 10.1109/ICOSP.2010.5657217. [19] R. Martin, "Noise power spectral density estimation based on optimal smoothing and minimum statistics," in IEEE Transactions on Speech and Audio Processing, vol. 9, no. 5, pp. 504-512, July 2001, doi: 10.1109/89.928915. S. M. Kim, "Hearing Aid Speech Enhancement Using Phase Diff erence-Controlled Dual-Microphone Generalized Sidelobe Canceller," in IEEE Access, vol. 7, pp. 130663-130671, 2019, doi: 10.1109/ACCESS.2019.2940047. [20] SNRVAD. [Online]. Available: https://labrosa.ee.columbia.edu/projects/snreval/. [21] C. Kim and R.M. Stern, "Robust signal-to-noise ratio estimation based on waveform amplitude distribution analysis," Proc. Interspeech 2008, 2598-2601, doi: 10.21437/Interspeech.2008-644. [22] A. W. Rix, J. G. Beerends, M. P. Hollier and A. P. Hekstra, "Perceptual evaluation of speech quality - A new method for speech quality assessment," Proc. Int. Conf. Acoust., Speech Signal Process. (ICASSP), 2001, pp. 749 752.