`RTL898_1021-0001
`
`Realtek 923 Ex. 1021
`Realtek 898 Ex. 1021
`
`
`
`
`
`FIG. I. The experimental setting,
`
`or postprocessing of two microphone signals by an adaptive
`bcamfonner.
`
`A block diagram of the adaptive beamforrner used is
`shown in Fig. 2. The sum and the difference of the micro-
`phone signals is calculated first. The difference signal, which
`contains mainly noise, drives the adaptive filter which calcu-
`lates an estimate of the noise in the sum signal.
`The coefficients of the FIR-structured adaptive filter are
`updated continuously in real
`time by a least-mean-squares
`ELMS} adaptation algorithm.“ The filter has a size of SOD
`coefficients {SD ms). The adaptation step size was chosen to
`be ill of the size which leads to instability of the adaptation
`algorithm, resulting in an adaptation time constant of [L125
`s. In order to maximize the improvement in SNR, the sum
`signal was delayed by 15% of the length of the adaptive filter
`(12.5 ms)?
`Whenever a strong desired signal arrives at the micro-
`phones, the adaptation of the filter is stopped. This is useful
`in order to prevent the adaptive beamfonner from canceling
`those parts of the desired signal, which do not arrive with th'e
`same phase and amplitude at both microphones. The idea of
`an adaptation inhibition was introduced by Crreenherg and
`Zurelgfi and Van Cortipernollefi The method of desired—signal
`detection used in this investigation is described and evatu-
`ated by Kornpisi The variances of the sum and difference
`signals are compared in an exponential window with a time
`constant of 10 ms. The filter adaptation is stopped, whenever
`the variance of the sum signal
`is greater than 1.5 times the
`variance of the -ditferenee signal.
`The test signals were presented by headphones to the
`normal hearing subjects. In order to allow the hearing aid
`users to use their own hearing aids during the experiments,
`
`
`
`FIG. 2. Bloc]; diagram of the adaptive hearnforrncr including a desired-
`signni-detectiou scheme.
`
`1911
`
`J. Ptooust. Soc. Am., vol. 95. No. til. September 1994
`
`RTL923_1021-0002
`RTL898_1021-0002
`RTL923_1021 -0002
`
`
`
`the test signals were presented through a loudspeaker in El
`separate and silent but not sound proof room to this group.
`The estimated direct-to-reverherant ratio of this presentation
`was 3.6 dB. The fieqaency response of the overall system is
`a complex function of the direction of incidence of the
`acoustic signal. ‘This is not only due to the characteristics of
`the directional microphones, but also an effect of the head
`shadow on any microphone attached to a head-sized object”
`and of the signal processing used. When adaptive beamform—
`ing is active,
`the frequency response may even vary with
`time. The restdting differences between the frequency re-
`sponses of the tested procedures were found to be small in
`informal listening tests. They could be compensated by using
`fixed filters, however one filter can compensate only for one
`single direction of incidence and, due to reverberation, for
`one acoustic environment. In order to avoid the inherent ar-
`
`bitrariness of this approach and to facilitate comparison with
`earlier studies using adaptive beamformersf” no fixed filters
`were used.
`
`ll. EXPERIMENTS Al-ID HESULT5
`
`intelligibility tests were performed with nine normal-
`hearing and six l-.earEng—impaired volunteers. To qualify for
`the experiments, normal hearing subjects were required to
`have no known hearing disorder and no lower hearing
`thresholds than 20 dB in the range 125 1-12 to E Id-lz. The
`hearing"-iropaircd subjects were all regular hearing aid users
`with gpre sensorineural hearing loss and used their hearing
`aid throughout all experiments described in this test. For
`both groups of volunteers the intelligibility of a desired sig-
`nal was compared in the
`three
`following microphone!
`processing conditions:
`(fl adaptive beamforrner with omnidirectional micro-
`phones;
`{ii} small microphone array with two directional micro-
`phones:
`{iii} adaptive beamformer with directional microphones.
`in condition {ii} the outputs of the two directional micro-
`phones at the ears of the dummy head were simply added. in
`addition, for the nonnal-hearing subjects l.l'tf: oondition
`[iv] stercophonic presentation of the unprocessed signals
`recorded by the omnitlireelional microphones; was tested in
`order to obtain a comparison with the binaural processing of
`the normal hearing volunteers.
`The desired signal emitted by the front loudspeaker were
`test words from a Gennan minimal pair test.“ In this test, the
`intelligibility of vowels and consonants is tested separately.
`A test word is presented together with a multiple choice of
`four possible answers which differ in one phoocmfi 0|1i!r'.' for
`example “I.ebe," “Liehe," “Lohe," and *‘Labe." The raw
`intelligibility scores fm, were chance-level corrected for the
`effect of guessing by the formula
`
`'rcorrertetfr=“'ra-ti.-'_CiiI'|iii_Cl1
`
`ll)
`
`with C=fl.25.
`
`The noise emitted by the loudspeaker to the right of the
`dummy head, was a speech-spectrum shaped and randomly
`amplitude-modulated noise introduced by Fastl.” For every
`test situation, two difierertt SNR's were tested. This set up
`
`M. I-(on-ipls and N. Dllllerz Letters to the Editor
`
`1911
`
`
`
`rraagtinyttstataeateuuearaaaa
`
`mg?
`
`‘I
`
`uaflflaflflfiflflfl
`
`-lfifl -10$ oi
`Geetlrnlla
`
`-lltl -IDS -GE Elli
`‘lltltda
`
`llll
`
`attentive ltaarntorrner with dlrattllttrtal microphones
`Adaptive bearrtlotrnar with omnidlractlonal microphones
`tttto'ophona
`rttlhtwodlraesenalrrrtetoohottaa
`Btrtu.rIiotIeerm'tt-hiorttstfltorru-tldtttetaiorralnitaoptinoea
`
`I-'16.}. Results at the intelligibility teats toeeoosoaants and vnwelutthree
`different signal-to-noise ratios. Llnfilled symbols are used to represent the
`results from experiment.-. with nine normal hearing volunteers [4-Sll test
`words per data point]; filled symbols denote the resttits frottt ear-erimenls
`with six hearing-irnpalred volunteers lavlllft tt words per data. point].
`
`allows the assessment of the processing schemes at different
`5NR's rather than the estimation of an average improvement
`in SNR. The SHE was defined by the levels of two test
`signals, corresponding to tlte average noise level and the av-
`erage level of all test words used [i.e.. for the consortant and
`the vowel teat] respectively. as measured above the dummy
`head. Fifty test words were presented to each volunteer for
`each combination of processing conditions. vowellconsonant
`test. and SNR. The test sequence was systematically varied
`in order to eliminate possible effects of training or fatigue.
`The noise signal was always present well in advance -of
`the fi'l.'SlI test word. ‘With its short adaptation time constant of
`[L125 s the beantforrner was in an adapted steady state
`throughout all experiments.
`Figure 3 shows the results of the intelligibility tests. The
`intelligibility for hearing aid users [filled symbols} is inher-
`ently lower than for nm:n:tal hearing volunteers {unfilled sym-
`bolsl. and therefore the average SHE used was chosen 5 dB
`higher for the fonner group. Generally. the adaptive beatn-
`former alone. i.e.. without directional microphones. already
`produces a more intelligible output signal
`than the simple
`rwo—direcn‘onal-microphone array. Tile best intelligibility for
`both groups of volunteers and all Si‘-lll‘s is achieved by the
`combination of directional microphones with adaptive beam-
`forming. The binaural processing of the normal hearing vol-
`unteers shows similar results as the microphone array. Since
`the binaural processing capabilities are also affected in most
`sensorineurally hearing impaired persons. still lower intelli-
`gibility scores would have been expected for this group.
`However this situation was excluded Erorn the test sequence
`because of the difficulties associated with a binaural presen-
`tation of signals to users of a single hearing aid. Subjective
`signal quality. as evaluated in informal listening tests. corre-
`sponds roughly to the intelligibility scores found.
`To evaluate the statistical significance of the improve-
`
`merits by the combination of directional microphone with
`adaptive beantfonning. a Wilcoxnn pair-«difference test was
`used. For all combinations of nonnal-hearing subjects;
`heatir1g—aid users. consonantalvowels and all SI'~lR's the in-
`telligibiiity obtained with the combination of directional mi-
`crophoncs and adaptive beamforming was compared to every
`other nticrophorter‘processing condition tested. The improve-
`meal was significant on a 593 level for all but three tests. the
`exceptions being consonants at —l[l dB for nonnal hearing
`volunteers with microphone array; consonants and vowels at
`—lll dB for hearing aid users with the adaptive bearnlormer
`alone. Although all volunteers did profit from the combina-
`tion of directional microphones with adaptive beamfonning.
`in the majority of all
`the conditions tested. they did so at
`different absolute levels of intelligibility. The standard devia-
`tions of the individual data points in Fig. 3 range from 3.1%
`to 13.49.‘: and are omitted for clarity.
`
`III. DIISCUSSIDH
`
`Experiments were perfonned to estimate the benefit of a
`combined directional mierophooeiadaptive heamfonner ap-
`proach for noise reduction for hearing aids. Normal hearing
`volunteers as well as treating aid users were tested at differ-
`ent SNR'a The intelligibility of vowels and consonants was
`measured separately. The results suggest that the combina-
`tion of directional microphones and adaptive beantfont-ting
`can improve the intelligibility of speech in noise sigrtlllcantly
`more than any one of these two approaches by itself. or- the
`binaural processing of normal hearing subjects. The direc-
`tional microphones also improve the reliability of the target
`signal detection. which results in a reduced cancellation of
`the desired signal at high Sl"iR’s. This might eitplain the
`different slopes for the adaptive beamformer with directional
`and omnidirectional microphones between —1I.'t and -5 dB
`in Fig. 3. The errperimenrs support our underlying hypothesis
`that combining directional microphones with an adaptive
`beamiormer is a useful and promising approach.
`To rate the relevance of these results for future practical
`hearing aids. several factors which have not been varied in
`the simple experiment deserve closer examination. The per-
`formance of the combined noise reduction scheme proposed
`depends most probably mainly on the following factors: {a}
`the direct—to-revert:-erant
`ratio of the signal- and noise
`sources. {hi the number. and let the placement of the noise
`sources The placement of the listener in a given environ-
`ment le.g.. close to it walll may also have some limited in-
`fluence. Since the performance of both the adaptive beam-
`lorrner and directional microphones increases at high direct-
`to-reverberant ratios. it can be expected that the combined
`noise reduction scheme behaves similarly. This suggests that
`hearing aid users ought profit most of the combined ap-
`proach there. where they are most affected. i.e.. in the vicin-
`ity of noise sources.
`Additional noise sources have no impact on the direc-
`tional microphones hut they can reduce the performance of
`the adaptive beamfonner substantially. However this is only
`critical when several noise sources with similar levels and
`spectra are present.”
`
`1912
`
`.l. Aoouat. 50¢. .ll.rn.. Vol. 95. ND. 3-. September 1994
`
`lvl. Komnla and N. Dllliar: Lahore to the Editor
`
`lstz
`
`RTL923_1021-0003
`RTL898_1021-0003
`RTL923_1021—0003
`
`
`
`
`
`As 10 tha direction of incidence of the. noise signal.
`adaptive lflcnoiionning, and dirnclional mimopbunos comple-
`ment
`I.-.a-ch other favorably. "The opening angle for desirad
`signal: is narrow for Ila: adapliw: beamftnmcr none.‘ |:Iu1
`do: tnthesymmctry with:-napocttotheaaraxis the from-
`to-lnck disuimixiation is poor. Dirocliornl roiu-opilnnta, in
`oonlrasl. show I broad opening magic but an reliably distin-
`guifl: bernraan the from and back directions.
`Thea: oonaitioratioos suggest I]aa.t an impnwtmeiil in in-
`tciligii:-iliigr
`by the combined directional micsophonflmi
`adaptive. boamforming apptoacli can also be expected in
`more realistic than tho givan experimental situation.
`
`AGKHDWLEDGHEME
`
`This work was supponnd by tin: Swiss Naiionul Ro~
`soarcii Foundation. Grant No. 4013-10364 and H1: Digital
`Speech PI'oocssil1g,De]:arln1:::tofA9oo1n'Ii:::lIl1d.We
`would lib: to thank the Phonal: Company for tho diroclzimul
`microphones and Dr. Wai-Kong Lai for his assistance in pre-
`paring this 14:1.
`
`'1. 5. Lirn and A. V. D-ppI:Ihr;irI1,"EI'HlI11-I3nI'III!1Il um: Eanniwlcith Compton-
`ainn oi’ Noisy Spcodz." Proccnd. I i‘.'il2I.1Sii6-IISIH llil'19l.
`Io. Hocllcr. -oi.-miumx microphone nu.-mg Mas: A. no yen mp-art."
`I-Int lnstmm. 31'-{ll}.
`l5—3!l ilflil.
`‘w. sausc. “lnnluvcnlal arspnu-. Inmnqjbilny In Hahn.“ mo. Ihcais.
`Dali‘! ilnimiiiy o=l‘To:hno|om'. Dolfi. Th: I*iol|I:I'iIn1Iai19!iDi.
`‘T. Boolorlilllcl. "Spatial: In.|Iil:iI:i:I1 it lliiiiol III‘S "
`I'|1.l.‘I.fl:n7sIIIl1el.iI'i'nIc1Ii'.l'_\I't:r!iIII[tIIlI.iciL Gurrolny {WM}.
`if‘. H. l'cI‘.orIl.'n. N. I. Dolilch, W. H. lubinowila. and P. M. Zanelt. ‘Huh
`Iin:lu'npiIon:I!IpLiv':br4.:nf:I1ui1sgicrlnJerfnI'=no= I'nim:I.ionio|1oIring
`Iilil." J. R::i'alI.R1:I. Dcv.1Ili4}. 103-[ID {INT}.
`"J. E lirocnhag and F. lut.1'.IlI’¢L "E\'I1I:I.Illon dill Idlpivc |aca|nl'-orming
`Ilmhod fin he-arillg Ii-Iia."JJIuonr.Il.'l.. Soc. Am. I1. IHII-—if|'.-'-I5: {.1992}.
`TM. Konapia, "Der adaptive B¢Imi’orI'no1'. E'l|'|iIJIlii‘.IlI oinna Vorflhruin zor
`Story-.1-iuachunhu-dridcung Iii
`I-IErrgariIc." Docioral dimcnatlon Na.
`Walill II the Swiss Federal lnalilnln of '|'ln:|1mla-g_3r [E'I"i-ll. Zurich, Swit-
`mlnnd {I993}.
`'13. Winlrnut, J. n. Glam, .1. M. Macml, J. Kaunilz, I:.s.wnI1ams. R. H.
`Hllrn. I. ll. Iaidler. E. Dong. and E. C. Goolilin, "milpliuln No-inn can-
`nailing: Principles an:lappii.c|Ii|:n'as," Froc. IEEE-H. 16-9'2-—lTli5 [lii':'5iL
`‘D.
`‘aha cnmpemnnc. "I-twin; Aim Ua.in|. Iiinlurni Pmamlng Prin-
`niplu,“ Mu ouuhryngci. esmeth}-. Suppl. Iii, 1544 {mo}.
`‘M. liompil mi N. Dilicr. "5in'n.laIiq. Iannlflrr finclionl in I rwtlbcrlnl
`Iucmindldiqlannnclihuctirltylnilfld-IIIdIfl:fl'tdt."l.fi£nluI.
`Son. Am. 53, 2'.'rI9—21&’l' il'H.'ll.
`
`" P.LCnib-ra. “Btnrictlnng and kliiiucisa Aunndlag clan ar.Iu1nItmLnI:r-
`munen lI&r=nt:."D-ncmnl dim=rua$un.UnIvmhy ufluidn. S'nri::|:I'-
`Iancl {I583}.
`“I-L Full, “Ein Stfiugcliind Eu liil Spraclmflouaotridh d
`Noiao for Spooch mn1ummy,"AudIuia;.MmuAudioIug. mun. 26:11.
`2.13 11931}.
`
`1913
`
`.l.AnoI.IIi.'.3oc.Arn..Vol.H.Ho.3.SaplnmhnrIBB-I
`
`I.I.Ko11pIundN.BIIor:LIflIIuoIInEcIInr
`
`1913
`
`RTL923_1021-0004
`RTL898_1021-0004
`RTL923_1021 -0004