`
`Microphone Arrays
`
`Signal Processing
`Techniques and Applications
`
`Springer
`Berlin
`Heidelberg
`New York
`Barcelona
`Hongkong
`London
`Milan
`Pais
`Singapore
`Tokyo
`
`oltullEuBRArY
`
`Engineering
`http/Amrursp d nger.de/en gine/
`
`Springer
`
`&
`
`Jawbone's Exhibit No. 2004, IPR2023-01134
`Page 001
`
`
`
`Series Editors
`Prof. Dr.-Ing. Anrlo L.e,cnolx
`f ohann-Wolfgang-Goethe-Universitit
`Institut fiir angewandte Physik
`Robert-Mayer-Str. 2-4
`D-60325 Frankfurt
`
`Prof. Dr.-Ing.
`Ax.e,srlsros VrNers.axopoulos
`University of Toronto
`Dept. of Electrical and Computer Engineering
`l0 King:s College Road
`M5S 3G4 Toronto, Ontario
`Canada
`
`Editors
`Prof. Mrcrrlnr. BRANDsTEIN
`Harvard University,
`Div. of Eng. and Applied Scciences
`33 Ot'ord Strea
`MA 02138 Cambridge
`USA
`e-mail: msb@hrlharvarde du
`Dr. DlnnBn Wl,no
`Imperid College, Dept. of Electrical Engineering
`Extribition Road
`SW7 2AZ london
`GB
`e-mail: d,ward@ic.ac.uk
`
`ISBN 3-540-41953-5 Springer-Verlag Berlin Heidelberg New York
`
`Gip data applied for
`
`This work is subject to copyright. All rights are reserve4 whether the whole or part ofthe material is
`concerned, specifically the rights of translation, reprinting reuse of illustrations, recitation,
`broadcasting,-reproduition on microfilm or in other ways, and storage in data banks. Duplication of
`this publication or parts thereofis permitted only under the provisions ofthe German Copyrightlaw
`of September 9, 1965, in its current version, and permission for use must always be obtained from
`Springer-Verlag. Violations are liable for prosecution act under German CoPyritht taw.
`
`Springer-Verlag is a company in the BertelsmannSpringer publishing group
`http://www.springer.de
`@ Springer-Verlag Berlin Heidelberg NewYork 2001
`Printed in Germany
`
`The use ofgeneral descripive names, registered names, trademarks, etc. in this publication does not
`impln even in the absence of a specific statement, that such names are exemPt from the relevant
`protective laws and regulations and therefore free for general use.
`
`Typesetting: Camera-ready copy by authors
`Cover-Design: de'blik, Berlin
`SPIN: 10836055 6213020 5 43210 Printedonacid-freepaper
`
`Preface
`
`The study and implementation of microphone arrays originated over 20 yeaxs
`ago. Thanks to the research and ecperimental developments pursued to the
`present day, the field has maturd to the point that array-based technologr
`now has immediate applicability to a number of current systems and a rrast
`potential for the improvement of existing products and the creation of future
`devices.
`In putting this book together, oru goal was to provide, for the first time,
`a single complete reference on microphone arrays. We invited the top re
`searchers in the field to contribute articles addressing their specific topicft)
`of study. The reception we receivd from our colleagues was quite enthusi-
`astic and very encouraglng. There was the general consensus that a work
`of this kind was well overdue. The results provided in this collection cover
`the current state of the art in microphone array resea,rch, development, and
`technological application.
`This text is organized into four sections which roughly follow the major
`areas of microphone array research today. Parts I and II are primarily the'
`oretical in nature and emphasize the use of microphone arrays for speech
`enhancement and source localization, respectively. Part III presents a num-
`ber ofspeciffc applications ofarray-based technology. Part IV addresses some
`open questions and explores the future of the field.
`Pa,rt I concerns the problem of enhancing the speech signal acquired by
`an array of microphones. For a variety of applications, including human-
`computer interaction and hands-free telephony the goal is to allow users to
`roam unfettered in diverse environments while still providing a high quality
`speech signal and robustness against background noise, interfering sources,
`and reverberation effects. The use of microphone arrays gives one the oppor-
`tunity to exploit the fact that the source of the desired speech signal and the
`noise sources are physically separated in space. Conventional array process-
`ing techniques, typically developed for applications such as radar and sonar,
`were initially applid to the hands-free speech acquisition problem. However,
`the environment in which microphone arrays is used is significantly different
`from that ofconventional array applications. Firstlg the desired speech signal
`has an extremely wide bandwidth relative to its center frequency, meaning
`that conventional na,rrowband techniques are not suitable. Secondly, there
`
`Jawbone's Exhibit No. 2004, IPR2023-01134
`Page 002
`
`
`
`VI
`
`Preface
`
`is significant multipath interference caused by room reverberation. Finally,
`the speech source and noise signals may located close to the arraS meaning
`that the conventional far-field assumption is typically not valid. These dif-
`ferences (amongst others) have mea,nt that new array techniques have had
`to be formulated for microphone array applications. Chapter 1 describes the
`design of an a,rray whose spatial response does not change appreciably over
`a wide bandwidth. Such a design ensures that the spatial filtering performed
`by the a^rray is uniform across the entire bandwidth of the speech signal. The
`main problem with many array designs is that a very large physical array is
`required to obtain reasonable spatial resolution, especially at low frequencies.
`This problem is addressed in Chapter 2, which reviews secalled superdirec-
`tive arrays. These anays are designed to archieve spatial directivity that is
`significantly higher than a standard delay-and-sum bea,mformer. Chapter 3
`describes the use of a single-channel noise suppression filter on the output
`of a mioophone array. The design of such a post-filter typically requires in-
`formation about the correlation of the noise between different microphones.
`The spatial correlation functions for various directional microphones a,re in-
`vestigated in Chapter 4, which also describes the use of these functions in
`adaptive noise cancellation applications. Chapter 5 reviews adaptive tech-
`niques for microphone arrays, focusing on algorithms that are robust and
`perform well in real environments. Chapter 6 presents optimal spatial filter-
`ing algorithms based on the generalized singular-value decomposition. These
`techniques require a large number of computations, so the chapter presents
`techniques to reduce the computational complexity and thereby permit real-
`time implementation. Chapter 7 advocates a new approach that combines
`explicit modeling of the speech signal (a technique which is well-known in
`single-channel speech enha,ncement applications) with the spatial filtering af-
`forded by multi-channel array processing.
`Part II is devoted to the source localization problem. The ability to locate
`and track one or more speech sources is an essential requirement of micro.
`phone array systems. For speech enhancement applications, an accurate fix
`on the prima^ry talker, as well as knowledge of any interfering talkers or coher-
`ent noise sources, is necessary to effectively steer the arra5 enhancing a given
`source while simultaneously attenuating those deemed undesirable. Location
`data may be used as a guide for discriminating individual speakers in a multi-
`source scenario. With this information available, it would then be possible to
`automatically focus upon and follow a given source on an extended basis. Of
`particula,r interest lately is the application of the speaker location estimates
`for aiming a camera or series of cameras in a videoconferencing system. In
`this rega^rd, the automated localization information eliminates the need for a
`human or number of human camera operators. Several existing commercial
`products apply microphon+array technology in small-room environments to
`steer a robotic camera and frarne active talkers. Chapter 8 summarizes the
`various approaches which have been explored to accurately locate an individ-
`
`Preface VII
`
`ual in a practical acoustic environment. The emphasis is on precision in the
`face of adverse conditions, with an appropriate method presented in detail.
`Chapter 9 extends the problem to the case of multiple active sources. While
`again considering realistic environments, the issue is complicated by the pres-
`ence ofseveral talkers. Chapter 10 further generalizes the source localization
`scenario to include knowledge derived from non-acoustic sensor modalities.
`In this case both audio and video signals are effectively combined to track
`the motion of a talker.
`Part III of this texi details some specific applications of microphone array
`technology available today. Microphone arrays have been deployed for a vari-
`ety of practical applications thus far and their utility and presence in our daily
`lives is increasing rapidly. At one extreme are large aperture arrays with tens
`to hundreds of elements designed for large roorns, distant talkers, and adverse
`acoustic conditions. Examples include the two-dimensional, harmonic array
`installed in the main auditorium of Bell Laboratories, Murray Hill and the
`5l2-element Huge Microphone Array (HMA) developed at Brown University.
`While these systems provide tremendous functionality in the environments
`for which they are intended, small a^rrays consisting of just a handful (usu-
`ally 2 to 8) of microphones and encompassing only a few centimeters of space
`have become fa,r more common and a,ffordable. These systems are intended
`for sound capture in close-talking, low to moderate noise conditions (such
`as an individual dictating at a workstation or using a hands-free telephone
`in an automobile) and have exhibited a degree of effectiveness, especially
`when compa^red to their single microphone counterparts. The technology has
`developed to the point that microphone axrays a,re now available in off-the-
`shelf consumer electronic devices available for under $150. Because of their
`growing popularity and feasibility we have chosen to focus primarily on the
`issues associated with small-aperture devices. Chapter 11 addresses the in-
`corporation of multiple microphones into hearing aid devices. The ability of
`beamforming methods to reduce background noise and interference has been
`shown to dramatically improve the speech understanding of the hearing im-
`paired and to increase their overall satisfaction with the device. Chapter 12
`focuses on the case of a simple two-element array combined with postfiltering
`to achieve noise and echo reduction. The performance of this configuration
`is analyzed under realistic acoustic conditions and its utility is demonstrated
`for desktop conferencing and intercom applications. Chapter 13 is concerned
`with the problem of acoustic feedback inherent in full-duplex communica,
`tions involving loudspeakers and microphones. Existing single'channel echo
`cancellation methods are integrated within a beamforming context to achieve
`enhanced echo suppression. These results are applied to single' and multi-
`channel conferencing scenarios. Chapter 14 explores the use of microphone
`arrays for sound capture in automobiles. The issues of noise, interference, and
`echo cancellation specifically within the car environment are addressed and a
`particularly effective approach is detailed. Chapter 15 discusses the applica-
`
`Jawbone's Exhibit No. 2004, IPR2023-01134
`Page 003
`
`
`
`Contents
`
`3 3 6 7 7 8II
`
`11
`T2
`L2
`13
`16
`16
`
`19
`19
`20
`2t
`22
`23
`24
`24
`24
`26
`26
`30
`30
`32
`33
`33
`
`Part I. Speech Enhancement
`I Constant Directivity Beo"'for:ning
`Danen B. Ward, Rodney A. Kennedy, Robert C. Williomson
`1.1 Introduction..
`1.2 Problem Formulation
`1.3 Theoretical Solution
`1.3.1 Continuous sensor
`1.3.2 Bearn-shaping function
`1.4 Practical Tmplementation
`1.4.1 Dimension-reducing para,meterization
`1.4.2 Reference bea,m-shaping filter
`1.4.3 Sensor placement .
`1.4.4 Su--ary of implementation ........
`1.5 Exa^mples ,
`1.6 Conclusions ......
`References
`2 Superdirective Microphone Arrays
`Joerg Bitzer, K. Uwe Simmer
`2.1 Introduction......
`2.2 Evaluation of Bea,mformers
`2.2.L Atray-Gain .
`2.2.2 Beampattern. .
`2.2.3 Directivity. .
`2.2.4 F\ont-to'Back Ra,tio .
`2.2.5 White Noise Gain ...
`2.3 Design of Superdirective Bea,mformers . .. .. . .. .
`2.3.1 Delay-and-Sum Bearnformen
`2.3.2 Design for spherical isotropic noise.
`2.3.3 Design for Cylindrical Isotropic Noise
`2.3.4 Design for an Optimal Flont-teBack Ratio ....
`2.3.5 Design for Measured Noise Fields
`2.4 Extensions and Details
`2.4.1 Alternative Form
`
`Vm Preface
`
`tion of microphone anays to improve the performance of speech recognition
`systems in adverse conditions. Strategies for effectively coupling the acous-
`tic signal enhancements afforded through bea,mforming with existing speech
`recognition techniques axe presentd. A specific adaptation of a recognizer to
`function with an array is presented. Finally, Chapter 16 presents an overview
`of the problem of sepa,rating blind mixtures of acoustic signals recorded at a
`microphone anay. This represents a very new application for microphone ar-
`rays, and is a technique that is funda,mentally different to the spatial filtering
`approaches detailed in earlier chapters.
`In the final section of the book, Part IV presents enpert summaries of
`current open problems in the field, as well as personal views of what the future
`of microphone axray processing might hold. These summaries, presented in
`Chapters 17 and 18, describe both academically-oriented research problems,
`as well as industry-focused areas where microphone array research may be
`headed.
`The individual ctrapters that we selected for.the book were designed to
`be tutorial in nature with a specific emphasis on recent imFortant results.
`We hope the result is a text that will be of utility to a large audience, from
`the student or pracbicing engineer just approaching the field to the advanced
`researcher with multi-channel signal processing experience.
`
`Cambridge MA, USA
`Iondon, UK
`Ja,nua,ry 2001
`
`Michael Bronilstein
`Donen Ward
`
`Jawbone's Exhibit No. 2004, IPR2023-01134
`Page 004
`
`
`
`F\rture Directions in Microphone Array
`18
`Proc
`esslng
`
`Dirk Van Compernolle
`
`Katholieke Universiteit Leuven, Leuven, Belgium
`
`18.1 Lessons From the Past
`Antenna array processing has had long-standing impact on phased array
`radars, sonaxs and radio astronomy for several deca.des. The gigantic an-
`tenna arrays that were constructed for deep space observation must stand
`out as some of the most impressive engineering achievements of any disci-
`pline. success in these related fields of signal processing have without any
`doubt stimulated interest in microphone array processing. And these suc-
`cesses did not only generate interest, they did much more-they created high
`expectations. Another interest generating stimulus came from a very different
`field, i.e. the one of anatomy and physiology. Nature has endowed virtually
`all species with two ears. one good reason, of course, is that there is always
`a second as backup when one of the two fails. But at the same time we all
`know that our sense for orientation is helped considerably by the use of two
`ears instead of one and that it helps us understand each other in the midst
`of a noisy crowd.
`After 20 years of active research, however, we cannot craim that micro
`phone array processing has had the success many of us hoped for, and many
`will wonder when the great breakthrough in microphone array processing
`will finally come, if ever- Nevertheless, progress in computer technology has
`helped us in a big way. In the early days only anarog ,"L"*", of limited sig-
`nal processing complexity were possible. This was followed by early years of
`high cost DSP computations, where computational cost seemed to impede
`widespread use of lhe technology. Today we have a.ffordable DSps that a,llow
`us to implement all but the most complex schemes cheaply in digital signal
`processing technology in real-time. But this in itself was not enough. Apart
`from breaking through the computing bottleneck, our understanding ofthe
`problems at hand has significantly progressed, as witnessed in this book. Most
`of the results presented are from recent years and give new insight into both
`the potential and the limitations of microphone axray processing. However,
`too often the same problems that were considered too hard ten or twenty
`yeaxs ago are still set apart for 'future resea.rch'. Admitted weaknesses to
`proposed solutions a,re similar to the ones that we have been struggling with
`for a long time. Generally speaking we may say that -*y ptoposed solu-
`tions add to our understanding but lack robustness in order to make a bright
`future for themselves.
`
`Jawbone's Exhibit No. 2004, IPR2023-01134
`Page 005
`
`
`
`390 Van Compernolle
`
`So should we not ask ourselves if there is a fundamental issue with micre
`phone array processing? And my answer is 'nothing is fundamentally wrong'.
`Microphone array processing has only proven to be quite a bit harder than
`other previously successful array processing applications. We have known the
`problems from the beginning, but have underestimated the impact of some
`of them in reallife situations.
`The basic problems fall into a small number of categories: (i) the speech
`signal is broadband; (ii) in many practical situations the desired source is
`in a reverberant space in the nea^r or mid field, is moving, and cannot be
`assumed to be a point source; and (iii) the speech signal changes rapidly,
`it is intermittent, and shares many cha^racteristics with the competing and
`interfering signals.
`It is very difficult to tackle all these issues at once. It is especially difficult
`to come up with tractable mathematical models for this complex environ-
`ment. The result of this complex situation is that a lot of resea,rch efort
`has gone into, and continues to go into the search for optimal beamforming
`strategies that rely on extra assumptions and constraints. Sadly enough, this
`all too often leads to solutions that lack robustness when evaluated in a ra-
`riety of reai-life situations. It may be that a far field assumption is reguired,
`it may be that less reverberation wouid be sufficient, or it may be that a
`perfect predictive speech detector will bring the breakthrough. Surely these
`mathematical developments are relevant and give us a better understanding
`of broadband beamforming in general. Simultaneously we should admit to
`ourselves that robustness has been, and still is today, one of the main issues.
`The drive to achieve (mathematically) optimal solutions is a natural un-
`derpinning of our science and engineering nature. But is microphone array
`processing not too complex to be solved with optimal approaches? Should
`we not expect real breakthroughs to come from so-called robust solutions
`that a^re clea^rly sub-optimal for any given circumstance, but applicable in a
`relatively wide range of situations? Also is it not obvious that there will not
`be a single solution, but that we need quite difierent solutions depending on
`the ta^rget application(s)? These observations go hand in hand with one of
`the major problems that has faced microphone a^rrays since their debut: size
`and cost. A large size always seemed a must from the requirement of uniform
`broadband beamforming. Some of the first microphone arrays, especially the
`one constructed in the auditorium at Bell Labs, were magically impressive
`by their shear size and number of microphones. They were great fun as a
`research project. Also they resulted in functional solutions. At the same time
`the price of such systems seems exorbitant. Later on we saw many axrays on
`the order of, say, 1 meter. Any such design is still only applicable to a very
`limited number of applications such as conference rooms. In the majority
`of potential applications such a bulky design has no place. No industry has
`screamed more for tiny and low cost solutions than the hearing aid industry.
`Here spacing of a few cm a^re the ma:<imum and processing power is an or-
`
`18 F\ture Directions in Microphone Array Processing 391
`
`der of magnitude less than in desktop applications. In all these situations we
`should not be surprised about the small size and limited number of sensors
`(two) in human hearing. It is fax from optimal, but it works.
`
`18.2 A F\rture Focused on Applications
`
`If we ask ourselves what will the future bring for microphone array processing,
`we must envisage a range of widely differing solutions of different sizes and
`costs:
`In the sequel I anaiyze the potential of the most important market seg-
`ments and go looking for killer applications. If commercialization has not yet
`started, the question is of course what harnpered commercial introduction
`and when, if ever, will we see usage of microphone arrays in each of these
`application fields.
`
`18.2.1 Automotive
`
`If any 'killer application' exists for microphone a,rrays, then it should be voice
`input in the ca.r. It has all the right ingredients. Mobile telephony and speech
`recognition scream for hands-free voice input in a noisy environment. Signal-
`to-noise ratios obtained by single microphones are just not sufficient. Thus
`microphone axrays seem the logical solution. There are extra features that
`should help. The speakers inside a cax axe not mobile and their position is
`reasonably constant from session to session. Ultimately, a market potential
`of tens of millions of units per yeax should be commercially convincing. All of
`ihis should be sufficient for successful uptake of microphone axray technology,
`but is it?
`Today, penetration of microphone axrays in ca^rs is minimal, except for
`a few top brands that axe not all that noisy by themselves, and therefore
`have the least need for it. The major concern of car equipment manufac-
`turers and car manufacturers alike is cost: multiple microphones, multiple
`wires, extra DSP power requAed, etc. Every cent in every component counts
`when putting a cax together and microphone arrays have been judged as too
`expensive. Also, at least for the foreseeable future one should envisage that
`most microphones are mounted into existing cars, further complicating the
`story for arrays.
`Given this large cost concern I do not believe that large arrays spanning
`the entire ca^r will ever be viable. On the contrary, the ca^r is an ideal envi-
`ronment for a microphone array that behaves like a traditional directional
`microphone but with a slightly steerable bea^rn. Such an embedded axray can
`be mounted as any regular microphone by not so specialized technicians. I
`do believe that the development of better microphones for usage inside the
`ca^r will be a point of focus for microphone developers in the coming yea^rs.
`
`Jawbone's Exhibit No. 2004, IPR2023-01134
`Page 006
`
`
`
`392
`
`Van Compernolle
`
`18.2.2
`
`Desktop
`
`Cost has been the stumbling block as well for desktop microphone arrays
`in conjunction with speech recognition in the PC environment. People just
`do not like it if you tell them that the accuracy of a $50 speech recognition
`software package will drastically improve if they buy a $150 microphone array
`to go with it. It is still unclear if we will ever overcome the cost hurdle in this
`case. It may just be a question of whether large enough volumes will ever be
`reached such that current prices can be lowered drastically.
`Microphone arrays for the desktop have just started to appear on the
`market. The reviews so far are ambiguous. In quiet environments they work
`as well as any headset worn. So if you do not want to be physically hooked
`up to your computer, this is the way to go the reviews say. At the same time
`the reviews will warn you that the existing commercial array microphones do
`not work well in considerable noise, and that one should not move around.
`Current reviews unanimously advise a wireless headset if one needs to move
`around a lot.
`It seems therefore that current commercial implementations only solve a
`small part of the problem. All of the designs rely prima^rily on fixed beam-
`forming, most often with limited directionality adjustment. On top of this,
`some additional noise suppression may be used. The 'speech seeking' part
`seems to be insufficient in all of the produced axrays. Also the quality and
`speed oftracking is substandard. It just shows how great the robustness issue
`really is when bringing microphone anay technology to consumer products.
`All in all there is reason for optimism, however. Desktop arrays axe very
`pragmatic in their designs. These microphones are built for applications that
`use a PC screen or monitor, and they sit perfectly well on top of a monitor or
`attach to the front of it. overall size is limited to about 20 cm, all computing
`is done inside the array, a^nd the a^rray connects to other equipment just a^s
`any other microphone would do. We have come a very long way to bring
`prices down enough such that a single enclosure with multiple elements, A/D
`converters and a DSP can be made at prices competing with traditional high
`end microphones. And let us not forget that these are first generation devices
`and that volumes axe still very small.
`Given some more time, I believe that there is hope that microphone arrays
`will capture a paxt of this market. Who knows, 5 years from now microphone
`arrays may be standard equipment on laptop and desktop computers. There
`is also a chicken and egg situation here. A wider usage of speech recognition
`would put more pressure on ha,rdwa^re manufactures to include higher end
`microphones, including arrays. On the other hand, one of the main hurdles
`in improving performance and subsequent acceptance of speech recognition
`is the low quality audio input on most systems today.
`
`18 F\rture Directions in Microphone Array Processiug Bg3
`1E.2.3 Hearing Aids
`Hearing aids form a market by themselves. Restrictions on size and compu-
`tational power axe an order of magnitude more stringent than in other areas,
`leading to substantially different designs. Array sizes of 5 to 20 cm have been
`used in experiments with hea^ring aids, but have overall been met with disap-
`proval. Nevertheless, here we have also seen the introduction of a range of new
`multi-micophone based products in the last couple of years. Many of these
`products do not use classical axrays, but a combination of microphones with
`different characteristics, used as inputs to a noise suppression stage. Perhaps
`even more obvious than in the automotive or desktop case, the evolution is
`towa^rds an adaptive speech seeking and noise suppressing microphone. The
`distinction here between microphone technology and a,rray technology is not
`entirely clea^r (but that does not really matter).
`
`18.2.4 Teleconferencing
`Teleconferencing was for some time seem as one of the potential killer ap
`plications. But I think that this is no longer true. On the one hand, the
`expansion of the teleconferencing ma^rket seems to have come down to slow
`growth and we see nothing of the explosion that some had hoped for. There.
`fore, the hope for a massive market does not seem justified. Acoustic eeho
`cancellation is the crucial issue and it can not be solved by array process-
`ing. When using arrays, as with any multi-microphone input, the problem
`becomes significantly worse. Special microphone designs, including radial a.r-
`rays, have been constructed and will continue to play a role in this ma^rket.
`Large wall mounted microphone a^rrays, however, are unlikely to find their
`way into teleconferencing rooms in any big way.
`
`f8.2.5 Very Large Arrays
`Teleconferencing was one of the potential markets for large axrays. Another
`one is the virtual microphone in la^rge auditoria. However, this can not be
`considered a booming market either. Design and manufacturing of these a.r-
`rays is costly and a la^rge degree of optimization may be required from site
`to site, making the picture even v/orse. Hence la^rge microphone arrays are
`doomed to remain a niche market. They will certainly survive in high profile
`demonstration projects, and as a research topic they will carry on for many
`years to come. Another (quite niche) market for very large a^rrays exists in
`the acoustic monitoring industry.
`
`18.2.6 The Signal Subspace Approach - An Alternative to
`Spatial Filtering ?
`Finally, we should ask ourselves the question if we should not look for al-
`ternative solutions to plain spatial filtering. We may think in two directions:
`
`Jawbone's Exhibit No. 2004, IPR2023-01134
`Page 007
`
`
`
`394 Van Compernolle
`
`blind signal separation and signal subspace approaches. These techniques do
`not require sensitive geometric information about the array layout but work
`with any configuration.
`These techniques should result in higher configuration robustness. But at
`the same time they are computationally very demanding and, while making
`fewer assumptions about the layout, they make in general more assumptions
`about the signals. Practical implementations have not appeared so far, but
`demonstration results are often impressive. So we should keep an eye open for
`these techniques. It is unlikely we will find them in products in the coming
`years, but in later generation array processing techniques, they may become
`the standard way to go.
`
`f8.3 Final Remarks
`
`The near-term trend is in one direction: small a.rrays with few microphones
`and a high degree of robustness that behave as speech seeking, directional,
`and noise canceling microphones. Depending on the target application designs
`may vary from less than a 1 cm in diameter for the hearing aid market, over
`5 cm for the car, to a maximum of 20 cm for desktop. After all, human hearing
`does very well with two ears spaced about 20 cm apart. These designs will not
`reach maximal noise suppression in any theoretical sense. Their goal is clea"r:
`a few dB gain in signal-to-noise ratio across the board at a cost which is only
`marginally above that of other microphones. A market of several million units
`for such medium cost devices is realistic and therefore economically viable.
`Economic potential for large arrays is much more limited and will therefore
`remain a niche markel.
`
`Index
`
`acoustic echo cancellation, 272, 28L,
`308, 386
`acoustic model, 335, 337
`adaptation control, 286
`adaptive ber-former, 88, 163, 235, 309,
`338
`afine projection algorithm, 285
`AIC, 195
`array calibration, 310
`array gain, 21,46,293
`array geometry, 340
`articulation index theory, 232
`artificial neural network, 334
`automotive microphone array, 309, 391
`autoregressive modeling, 160, 191
`
`Bark spectral distortion, 148
`Baum-Welch algorithm, 335, 345
`beamforming, 3, 87, 288
`bgampattern, 6, 22
`blind signal sepa.ration, 325, 355
`blocking matrix, 34, 88, 29L
`
`cardioid microphone, 69, 257
`causal separation model, 358
`cepstral mean subtraction, 337
`cepstral prefi.ltering, 162
`close-talking microphone, 345, 355, 385
`clustering, 195
`cochlear implant, 230
`cocktail party effect, 355
`coefRcient-constrained adaptive filters,
`94
`coherence, 22, 63,143, L6t,244
`coherence matrix, 22,45, 47
`coherence of ca^r noise, 259
`coherence of offce noise, 259
`coherence of speech, 259
`
`coherence-based post-filter, 271
`coherent signal subspace method, 183
`computed tomography scanner, 256,
`266
`constarxt directivity bea-former, 4
`constrained minimum variance
`bea-former, 310
`constrained optimization, 309
`contrast functions, 364
`correlation matrix, 114, 160, 314
`correlation-based criteria, 366
`cross-correlation, 40, 166
`cross-correlation matrix, 114
`cross-power spectrum, 189, 309
`cross-power spectrum phase, 190, 338
`cumulant-ba.sed separation criteria, 366
`cylindrically isotropic noise field, 30, 73
`
`data association problem, 222
`decentralized Kalman filter, 212
`decorrelation system, 371
`delay-and-sum beatnformer, 26, 159,
`169,234,290, 339
`density modeling criteria, 362
`desktop microphone array, 392
`differential microphone, 67
`diffuse noise field, 23, 309, 338
`diffuse sound field, 49,257
`dipole microphone, 69
`direction of arrival, L62, L82,243,290,
`302
`directional gain, 384
`directivity index, 23, 44,293
`distributed multi-microphone, 348
`disturbed ha^rmonics model, 189
`double-talk, 65
`dual excitation speech model, 137
`dyna-ic time warping, 334
`
`Jawbone's Exhibit No. 2004, IPR2023-01134
`Page 008
`
`