A Different Phase-Based Improved Performance of Differential

Microphone Array

Quan Trong The

Digital Agriculture Cooperative, No. 15 Lane 2, Tho Thap, Dich Vong, Cau Giay, Hanoi, Viet Nam

Keywords:

Microphone Array, Different Phase, Differential Microphone Array, Dual-System, Speech Enhancement,

Noise Reduction, Background Noise.

Abstract:

Various speech applications such as speech separation, speech recognition, communication, teleconference,

hearing aids common use microphone array (MA) as front-end speech enhancement structure. By using the

spatial information, geometry of MA, the characteristic of recording environment, MA help overcome of

single-channel algorithm to suppress noise without speech distortion. A critical major component of MA

system is the beamforming technique, which forms beampattern to a certain target directional sound source

while alleviating different noise. Differential Microphone Array (DIF) is one of the most popular structure

is used in several speech application. DIF has a lot of advantages which own the capability of null-steering

beampattern to the noise source. In this article, the author proposes a method to enhance performance of DIF

for separating speaker in real dialogue conference. Experiment was illustrated in real situation and obtained

results show the effectiveness of the suggested method in comparison with the previous author’s work.

1 INTRODUCTION

In several speech communication systems, separation

target directional speaker of a mixture of interference,

noise and third-party speaker is a challenging prob-

lem especially in situation with complex surround-

ing noise and low SNR (signal-to-noise ratio) values.

In general, due to the above reason, the speech qual-

ity and speech intelligibility are often signiﬁcant de-

graded. The necessary of communication is need to

improve speech enhancement, even in scenario of low

SNR is an importance research. Single-channel ap-

proach can’t adapt all requirement, because of it uses

spectral subtraction method, which lead to speech dis-

tortion of the ﬁnal output. Therefore, MA technology

has been developed for dealing this problem. Nowa-

days, MA is used in almost speech applications, such

as speech recognition, mobile device, hearing aid,

teleconference, human-machine interface in order to

obtain an acceptable degree of speech quality from

any algorithms trying to mitigate background noise

and extracting desired speech. MA exploits the spa-

tial diversity, the priori of knowledge of the direction-

of-arrival (DOA), the properties of acoustic situation,

the characteristic of noise ﬁeld to achieve a beampat-

https://orcid.org/0000-0002-2456-9598

Figure 1: The complicated task of separating desired speech

source.

tern, which toward speech source and remove all of

noise.

Because of possibly changing recording scenario

and varying position of the noise source relative to

MA, DIF beamformer allows null-steering beampat-

tern to noise. In the previous work, the author sug-

gested an additive equalizer for enhancing separation

The, Q.

A Different Phase-Based Improved Performance of Differential Microphone Array.

DOI: 10.5220/0012009300003561

In Proceedings of the 5th Workshop for Young Scientists in Computer Science and Software Engineer ing (CSSE@SW 2022), pages 34-40

ISBN: 978-989-758-653-8; ISSN: 2975-9471

 2023 by SCITEPRESS – Science and Technology Publications, Lda. Under CC license (CC BY-NC-ND 4.0)

Figure 2: The advantage of utilizing the MA beamforming.

each talker when using DIF.

G. W. Elko (Elko, 1997, 2000; Teutsch and Elko,

2001) formed a cardioid beampattern when combined

signal with compact spaced microphones. Therefore,

the directivity index of MA was increased.

The designing with small arrays and an appro-

priate geometry and associated beamforming algo-

rithms, which can pass all through bandpass speech

signals still a challenging problem to the scholars

(Barfuss et al., 2017a,b). Among different several

types of MA, DMA is the most suitable architecture

to measure differential of the sound target desired

speech , and more appropriate for achieving high di-

versity, high noise reduction, high directivity index

and invariant-frequency beampattern toward the di-

rection of talker (De Sena et al., 2012; Buck, 2002).

Recently, many efforts are successful with ﬂexible

differential beamforming and robustness of DIF’s per-

formance are enhanced (Huang et al., 2020; Benesty

et al., 2019; Cohen et al., 2019).

An essential issue in differential beamforming is

steering ﬂexibility in any direction of signal or noise.

Linear DMAs do not have much ﬂexibility, due to the

beampattern varies with steering angle and the opti-

mum directivity factor occurs only in end-ﬁre case

DMA. A numerous works are attempted to improv-

ing the ﬂexibility of steering DMA’s beampattern. In

(Elko and Pong, 1997; Derkx and Janse, 2009), a

two-dimensions MA are utilized to steer the result-

ing beampattern in a certain number of directions.

First-order steerable DMAs (Wu et al., 2014; Wu and

Chen, 2016) are evaluated to construct a combina-

tion of monopole and dipoles using 4 microphones

square array. In (Benesty et al., 2015), uniform circu-

lar DMAs were performed to steering beampattern to

some wanted directions. Bernardini et al. (Bernar-

dini et al., 2017) suggested using DMA to steer

second-order with monopole and dipole. In (Huang

et al., 2017), a beamforming technique was evalu-

ated to design an approximately constant beampat-

tern, and has capable to steer between two directions

of the reference beams. In (Parra, 2005, 2006), a de-

signed method was suggested for forming invariant-

frequency beampattern for spherical array. In (Lai

et al., 2013), a broadband beampattern was presented

with arbitrary geometry of microphones for spherical

MA.

With a great obtained progress in designing steer-

able DMA’s beampattern with high directivity index,

good noise reduction, more robustness, and high ca-

pability of steering beampattern.

In this paper, the author continues to deal the

problem of separating talker when using the subspace

technique to reduce the remaining speech component

non-target talker at the output DIF beamformer.

The rest of this paper is organized as follow. The

next section presents model signal, advantage of DIF

and the previous author’s work. Section 3 introduce

how to use the information about phase error between

two microphones to form a post-ﬁltering, which based

on Wiener ﬁlter. Section 4 described experiments

with two speakers, and a comparison between the pro-

posed method and the previous work. Concluding re-

mark and future work is presented in Section 5.

2 DIFFERENTIAL MICROPHONE

ARRAY

We will consider the DMA2’s working in STFT do-

main. DMA2 allows obtaining high diversity, noise

reduction, directional beampattern to target speech.

DMA2 have super directivity, small size and can

null-steering beampattern to the direction of noise,

which makes it possible to achieve high performance,

even in reverberation, noisy environment, and suitable

for almost acoustic equipment. The obtained effec-

tiveness of DMA2 is suppressing complicated back-

ground noise and extracting the target speaker.

The representation of two noisy microphone ar-

rays X

( f ,k),X

( f ,k) in the STFT domain as:

( f ,k) =S( f , k)e

jΦ

(1)

( f ,k) =S( f , k)e

− jΦ

(2)

where f ,k is the index of frequency and current con-

sidered frame, Φ

= π f τ

cos(θ

),τ

= d/c,d is the

distance of two microphones, c = 343(m/s) is the

speech of sound propagation in the air, Θ is the

direction-of-arrival (DOA) of useful speech.

With a certain delay τ is added, we can obtain high

directivity beampattern toward two speakers. The

output signals, which are based on subtraction signal

A Different Phase-Based Improved Performance of Differential Microphone Array

Figure 3: The MA signal processing in the frequency domain.

can be expressed as:

1DIF

( f ,k) =

( f ,k) − X

( f ,k)e

− jωτ

(3)

= jS( f , k)sin(

ωτ

(cosθ +

)) (4)

2DIF

( f ,k) =

( f ,k) − X

( f ,k)e

− jωτ

(5)

= − jS( f , k)sin(

ωτ

(cosθ −

)) (6)

The directional two beampatterns, which received

from equation (3) - (6) are derived as:

( f ,θ) =|

( f ,k)

S( f , k)

| = |sin(

ωτ

(cosθ +

))| (7)

( f ,θ) =|

( f ,k)

S( f , k)

| = |sin(

ωτ

(cosθ −

))| (8)

Stolbov et al. (Stolbov et al., 2018) proposed uti-

lizing an equalizer for preserving the useful signal at

the low-frequency band. The equalizer can be ex-

pressed as the following equation:

( f ) =











6 0 Hz < f ≤ 200 (Hz)

sin(

)

200 Hz < f ≤ Fc

1 Fc < f ≤ 2*Fc

0 2*Fc < f

(9)

where Fc =

4∗τ

. A constant threshold 12(dB) is de-

termined for H

q( f ). The ﬁnal output signals are:

Figure 4: The beampatterns.

( f ,k) =Y

1DIF

( f ,k) ∗ H

( f ) (10)

( f ,k) =Y

2DIF

( f ,k) ∗ H

( f ) (11)

In many realistic scenario, the remaining noisy

component at the output signal often deteriorate the

speech quality of DIF beamformer. In the next sec-

tion, the author show an additive post-ﬁltering, which

based on the priori information of DMA2.

CSSE@SW 2022 - 5th Workshop for Young Scientists in Computer Science Software Engineering

3 THE POST-FILTERING

An additive post-ﬁltering, which based on the sub-

space technique is presented in ﬁgure 5.

Figure 5: The suggested post-ﬁltering for DMA2 to sepa-

rate useful signal.

In the STFT-domain, with the assumption as the

noise magnitudes are equal for two microphones. The

different phase θ

( f ) is deﬁned as:

( f ) = arg(X

( f ,k)) − arg(X

( f ,k)) − 2π f τ

(12)

A major information that the different phase is al-

ways in range [−π..π]. In the frames, which contain

the speech component, θ

( f ) tends to 0 and in the

noisy frame, θ

( f ) tends to π or π.

The author proposed a post-ﬁltering, which use

the a priori information θ

( f ) as:

PF( f ) = cos

( f )

(13)

At the frames, in which the desired target speaker

exits, PF( f ) equal approximately 1 and will save the

speech component. Thus in, at the other noisy frames,

PF( f ) close to 0 and remove the interference or back-

ground noise.

At the ﬁnal, the received output signal are deter-

mined as:

1out

( f ,k) =Y

( f ,k) ∗ PF( f ) (14)

2out

( f ,k) =Y

( f ,k) ∗ PF( f ) (15)

In the next section, the author illustrated and ex-

periment to very the useful of the proposed post-

ﬁltering in compared with the previous author’s work

(Stolbov et al., 2018) by using a DMA2 to extract

the desired target talker while suppressing the inter-

ference.

4 EXPERIMENTS AND

DISCUSSION

The purpose of this experiment is enhancing the per-

formance, which based on the DIF beamformer. The

derived problem is alleviating the second talker’s

component while preserving target directional ﬁrst

talker. A degree of suppression second speaker’s

speech component is calculated to conﬁrm the

promising results. The scheme of experiment is de-

picted in ﬁgure 6. DMA2 with the inter-microphone

distance is d = 5(cm). The desired speaker talk while

the interference at the other opposite direction.

For calculating the spectral power density, some

parameters were used for transformed into STFT do-

main: Hamming window, NFFT=512, overlap 50%,

the sampling frequency Fs = 16kHz, smoothing pa-

rameter α = 0.1. A DMA2 was used for recording

speech from target talker at the direction θ

= 0(deg)

in presence of unwanted interference, which stand at

the opposite direction θ

= 180(deg).

Figure 6: The evaluated recording scenarios with two

speakers.

The waveform of the received signal was illus-

trated in ﬁgure 7.

In ﬁgure 8, the obtained processed signal by the

previous author’s work (Stolbov et al., 2018) and the

proposed method are illustrated in ﬁgure 9.

From these ﬁgures, as we can see that, the remain-

ing of speech component of the second talker is re-

moved to 5(dB) in comparison with (Stolbov et al.,

2018). The effectiveness of the proposed technique

was conﬁrmed in suppressing the remaining speech

component of second talker while ensuring keep the

ﬁrst second’s speech. This is the major advantage in

comparison with the previous work (Stolbov et al.,

2018).

So in this paper, the author exploits the different

phase approach to forming an additive post-ﬁltering

to enhance the ﬁnal desired target speaker. As can

be observed, the enhancement of suggested method

A Different Phase-Based Improved Performance of Differential Microphone Array

Figure 7: The waveform of the capture microphone array signal.

Figure 8: The processed signal by (Stolbov et al., 2018).

Figure 9: The obtained processed signal by additive post-ﬁltering.

was more signiﬁcant realized than the work (Stolbov

et al., 2018). This technique can be applied into other

acoustic system.

DMA2 has the capability of impact, low com-

putation, easy implementation. These advantages

make DMA2 is commonly installed in almost acous-

tic equipments; therefore the demand of enhancing of

of high directivity, more robustness still exits. The

author use the available information about difference

phase to improve the previous work.

Separation of target speaker in complex environ-

ment, in which the third-party talker or background

noise existed, still a challenge to all scholars. The

use of the characteristics between two microphones

is the advantage of MA to improve the speech en-

hancement. Different phase is one of the most inter-

esting object for studying to improve the performance

of DMA2.

5 CONCLUSION

MA beamforming are now common installed in var-

ious range of acoustic applications such as cel-

lular phones, teleconferencing, speech recognition,

robotics, smart device, telephone hands-free. A

challenging task is perfectly separating speech each

speaker, which requires using model a suitable DMA.

CSSE@SW 2022 - 5th Workshop for Young Scientists in Computer Science Software Engineering

In this paper, the author suggested exploit the charac-

teristic subspace of microphone array signals to im-

prove and increase the robustness of DMA system

than the previous work. A knowledge of the power

target speech source used for post-ﬁltering to suppress

the non-target different speaker without speech distor-

tion of target source. Numerical result was veriﬁed in

real scenario and conﬁrm the ability of the proposed

method. Phase error approach can be further studied

in future for installing into multi-microphone system.

In the future, the author will investigate the property

of environment to enhance the performance of MA

algorithms.

ACKNOWLEDGEMENTS

This research was supported supported by Digital

Agriculture Cooperative. The author thank our col-

leagues from Digital Agriculture Cooperative, who

provided insight and expertise that greatly assisted the

research.

REFERENCES

Barfuss, H., Bachmann, M., Buerger, M., Schneider, M.,

and Kellermann, W. (2017a). Design of robust two-

dimensional polynomial beamformers as a convex op-

timization problem with application to robot audi-

tion. In 2017 IEEE Workshop on Applications of Sig-

nal Processing to Audio and Acoustics (WASPAA),

pages 106–110. https://doi.org/10.1109/WASPAA.

2017.8170004.

Barfuss, H., Buerger, M., Podschus, J., and Kellermann, W.

(2017b). HRTF-based two-dimensional robust least-

squares frequency-invariant beamformer design for

robot audition. In 2017 Hands-free Speech Communi-

cations and Microphone Arrays (HSCMA), pages 56–

60. https://doi.org/10.1109/HSCMA.2017.7895561.

Benesty, J., Chen, J., and Cohen, I. (2015). Design of Cir-

cular Differential Microphone Arrays, volume 12 of

Springer Topics in Signal Processing. Springer Cham.

https://doi.org/10.1007/978-3-319-14842-7.

Benesty, J., Cohen, I., and Chen, J. (2019). Array Process-

ing: Kronecker Product Beamforming, volume 18 of

Springer Topics in Signal Processing. Springer Cham.

https://doi.org/10.1007/978-3-030-15600-8.

Bernardini, A., D’Aria, M., Sannino, R., and Sarti, A.

(2017). Efﬁcient Continuous Beam Steering for Pla-

nar Arrays of Differential Microphones. IEEE Signal

Processing Letters, 24(6):794–798. https://doi.org/10.

1109/LSP.2017.2695082.

Buck, M. (2002). Aspects of ﬁrst-order differential mi-

crophone arrays in the presence of sensor imper-

fections. European Transactions on Telecommuni-

cations, 13(2):115–122. https://doi.org/10.1002/ett.

4460130206.

Cohen, I., Benesty, J., and Chen, J. (2019). Differential

Kronecker Product Beamforming. IEEE/ACM Trans-

actions on Audio, Speech, and Language Processing,

27(5):892–902. https://doi.org/10.1109/TASLP.2019.

2895241.

De Sena, E., Hacihabiboglu, H., and Cvetkovic, Z. (2012).

On the Design and Implementation of Higher Order

Differential Microphones. IEEE Transactions on Au-

dio, Speech, and Language Processing, 20(1):162–

174. https://doi.org/10.1109/TASL.2011.2159204.

Derkx, R. M. M. and Janse, K. (2009). Theoretical Anal-

ysis of a First-Order Azimuth-Steerable Superdirec-

tive Microphone Array. IEEE Transactions on Au-

dio, Speech, and Language Processing, 17(1):150–

162. https://doi.org/10.1109/TASL.2008.2006583.

Elko, G. W. (1997). Adaptive noise cancellation with di-

rectional microphones. In Proceedings of 1997 Work-

shop on Applications of Signal Processing to Audio

and Acoustics, pages 4 pp.–. https://doi.org/10.1109/

ASPAA.1997.625628.

Elko, G. W. (2000). Superdirectional Microphone Ar-

rays. In Gay, S. L. and Benesty, J., editors, Acous-

tic Signal Processing for Telecommunication, vol-

ume 551 of The Springer International Series in En-

gineering and Computer Science, pages 181–237.

Springer US, Boston, MA. https://doi.org/10.1007/

978-1-4419-8644-3

10.

Elko, G. W. and Pong, A.-T. N. (1997). A steerable and

variable ﬁrst-order differential microphone array. In

1997 IEEE International Conference on Acoustics,

Speech, and Signal Processing, volume 1, pages 223–

226. https://doi.org/10.1109/ICASSP.1997.599609.

Huang, G., Benesty, J., and Chen, J. (2017). On the

Design of Frequency-Invariant Beampatterns With

Uniform Circular Microphone Arrays. IEEE/ACM

Transactions on Audio, Speech, and Language Pro-

cessing, 25(5):1140–1153. https://doi.org/10.1109/

TASLP.2017.2689681.

Huang, G., Chen, J., and Benesty, J. (2020). Design of pla-

nar differential microphone arrays with fractional or-

ders. IEEE/ACM Transactions on Audio, Speech, and

Language Processing, 28:116–130. https://doi.org/10.

1109/TASLP.2019.2949219.

Lai, C. C., Nordholm, S., and Leung, Y. H. (2013). Design

of Steerable Spherical Broadband Beamformers With

Flexible Sensor Conﬁgurations. IEEE Transactions on

Audio, Speech, and Language Processing, 21(2):427–

438. https://doi.org/10.1109/TASL.2012.2219527.

Parra, L. C. (2005). Least squares frequency-invariant

beamforming. In IEEE Workshop on Applica-

tions of Signal Processing to Audio and Acous-

tics, 2005., pages 102–105. https://doi.org/10.1109/

ASPAA.2005.1540179.

Parra, L. C. (2006). Steerable frequency-invariant beam-

forming for arbitrary arrays. The Journal of the Acous-

tical Society of America, 119(6):3839–3847. https:

//doi.org/10.1121/1.2197606.

Stolbov, M., Tatarnikova, M., and The, Q. T. (2018).

A Different Phase-Based Improved Performance of Differential Microphone Array

Using Dual-Element Microphone Arrays for Auto-

matic Keyword Recognition. In Karpov, A., Jokisch,

O., and Potapova, R., editors, Speech and Com-

puter, volume 11096 of Lecture Notes in Com-

puter Science book series, pages 667–675, Cham.

Springer International Publishing. https://doi.org/10.

1007/978-3-319-99579-3 68.

Teutsch, H. and Elko, G. W. (2001). First- and second order

adaptive differential microphone arrays. In Seventh

International Workshop on Acoustic Echo and Noise

Control, IWAENC 2001, Darmstadt. https://www.

iwaenc.org/proceedings/2001/main/data/teutsch.pdf.

Wu, X. and Chen, H. (2016). Directivity Factors of the

First-Order Steerable Differential Array With Micro-

phone Mismatches: Deterministic and Worst-Case

Analysis. IEEE/ACM Transactions on Audio, Speech,

and Language Processing, 24(2):300–315. https://doi.

org/10.1109/TASLP.2015.2506269.

Wu, X., Chen, H., Zhou, J., and Guo, T. (2014). Study

of the Mainlobe Misorientation of the First-Order

Steerable Differential Array in the Presence of Micro-

phone Gain and Phase Errors. IEEE Signal Processing

Letters, 21(6):667–671. https://doi.org/10.1109/LSP.

2014.2312729.

CSSE@SW 2022 - 5th Workshop for Young Scientists in Computer Science Software Engineering