Interactions of Hearing Aid Compression Release Time and Fitting

Abstract

The effects of the interaction of compression release time and prescribed gain on run-ning speech processed through a hearing aid on KEMAR was investigated. A digitalinstrument was programmed to fit a mild to moderate sloping hearing loss using probemicrophone measures to reach targets prescribed by NAL-NL1, DSL I/O, FIG.6 or ASA2pwith release times of 40 and 640 ms for each condition. Recordings were made throughKEMAR and analyzed to determine the long-term-average-speech spectra, consonant-to-vowel ratios and the RMS amplitude of 32 phonemic units. Aided and unaided resultswere compared. Within each prescriptive formula, changes in release time affected allof the speech measures subsequent to programming the instrument to a static-com-posite signal. The short release-time condition produced the greatest alteration to thespeech signal. Release time may need consideration when fitting hearing aids to targetgain prescriptions.

Key Words: Compression, compression attack time, compression release time, com-pression ratio, compression threshold, wide dynamic range compression, prescriptiveformula, Knowles Electronics Manikin for Acoustic Research, speech envelope.

Abbreviations: ASA2p = Adaptive Speech Alignment, B&K = Brüel and Kjær, CVR =Consonant-Vowel Amplitude Ratio, DSL I/O = Desired Sensation Level Input / Output,KEMAR = Knowles Electronics Manikin for Acoustic Research, LTASS = Long TermAverage Speech Spectrum, NAL-NL1 = National Acoustic Laboratories nonlinear pro-cedure Version 1, REIG = Real Ear Insertion Gain, REUG = Real Ear Unaided Gain,RMS = Root Mean Square, SPL = Sound Pressure Level, VCV = Vowel-Consonant-Vowel Stimuli, WDRC = Wide Dynamic Range Compression

Sumario:

Se investigaron los efectos de la integración del tiempo de liberación de la compresión(compression release time) y de la ganancia prescrita, utilizando lenguaje corrido através de un auxiliar auditivo en el KEMAR. Un dispositivo digital fue programado paraamplificar una hipoacusia leve a moderada con pendiente progresiva, utilizando medi-das de un micrófono de prueba que alcanzaran los objetivos indicados en NAL-NL1,DSL I/O, FIG.6, ASA2P, con tiempos de liberación de 40 a 640 ms para cada condición.Los registros se realizaron a través del KEMAR y fueron analizados para determinarlos espectros promedio a largo plazo del lenguaje, las tasa consonante/vocal y la ampli-tud RMS de 32 unidades fonémicas. Se compararon los resultados con y sin amplificación.Dentro de cada fórmula de prescripción, los cambios en el tiempo de liberación afec-taron todas las medidas de lenguaje, luego de programar el instrumento para una señalestática-compuesta. La condición de tiempo de liberación corto produjo la mayor alteraciónde la señal de lenguaje. El tiempo de liberación debería ser tomado en cuenta cuandose ajuste un auxiliar auditivo para alcanzar una ganancia de prescripción.

Interactions of Hearing Aid CompressionRelease Time and Fitting Formula: Effects onSpeech AcousticsJohn C. Ellison*Frances P. Harris*Thomas Muller*

*University of Arizona, Department of Speech and Hearing Sciences, Tucson, AZReprint Requests: Frances P. Harris, Speech & Hearing Sciences, 1131 E. 2nd St., Tucson, AZ 85721 Phone: (520) 626-7530 Email: [email protected]

Portions of this paper were presented at the American Academy of Audiology 2002 convention; April 20, 2002; PennsylvaniaConvention Center, One Convention Center Pl., 1101 Arch St., Philadelphia, PA 19107

59

Although selection of an appropriate fit-ting formula and of compression char-acteristics are both routinely consid-

ered when fitting a hearing aid to an individ-ual, the interactions of the two are often not.Gain as a function of frequency may be estab-lished by one of a number of prescriptive for-mulas, none of which include compressionrelease time in their calculations. Yet, changesin compression release time are known to influ-ence speech. It would seem appropriate toknow how gain may be influenced by changesin compression release time and what effectssuch changes might have on speech. Suchinteractions are the topic of this investigation.

Compression can alter speech characteris-tics in several ways. Compression-release timecan affect the temporal amplitude-modulationenvelope of a speech signal (Van Tasell, 1993).The effect of long release times on the modula-tion envelope is minimal relative to shortrelease times. Kuk (1998) claims that longrelease times maintain the short-term intensi-ty relationships in speech. Short release times,those equivalent to or shorter than the dura-tion of a syllable, allow the compression to varyduring the course of a word. As such, widedynamic range compression (WDRC) instru-ments with short release times compress high-level low-frequency segments (i.e., vowels) andrelease fast enough to provide gain to low-levelhigh-frequency segments (i.e., consonants)(Van Tasell, 1993). Relative to the input, theamplitude envelope of speech will then besmoothed or distorted and the consonant-vowelamplitude ratio (CVR) increased (i.e., conso-nant amplitude is increased) at the output.

Rosen (1992) described the speech enve-lope as amplitude fluctuations in speechbetween 2 and 50 Hz that provide segmentalcues for manner of articulation, voicing, vowelidentity, and prosody. Because individuals withprofound hearing impairments typically havedegraded auditory frequency selectivity, Rosen

implies that these individuals rely greatlyupon temporal cues, such as the amplitudeenvelope of speech. In support of this, severalinvestigators have indicated a positive correla-tion between the amount of access to speechenvelope cues and performance on word or syl-lable recognition tasks for subjects with pro-found hearing loss (Erber, 1972; Boothroyd etal, 1988).

Several investigators (Gordon-Salant, 1986,1987; Freyman and Nerbonne, 1989; Preves etal, 1991; Balakrishnan et al, 1996; Souza andTurner, 1996; Hickson and Byrne, 1997;Sammeth et al, 1999; Smith and Levitt, 1999;Souza, 2000;) have studied how the amplitudeenvelope or CVR of speech affects the intelligi-bility of speech units (i.e., items no longer than aword); however, results have been conflicting.Outcomes obtained using longer stimuli aremore consistent. Souza and Kitch (2001) stud-ied the effects of altering amplitude envelopecues on sentence identification while controllingfor audibility for normal-hearing and hearing-impaired adults each divided into groups of eld-erly and young adults. The authors determinedthat recognition scores became poorer as com-pression ratios increased from 1:1 to 2:1 andfrom 2:1 to 5:1, thus demonstrating a negativecorrelation between performance and the degreeto which amplitude envelope cues were altered.Van Tasell and Trine (1996) conducted threeexperiments to study the degree to which theamplitude envelope contributes to VCV disylla-ble and sentence recognition in normal-hearingsubjects. The authors found that severe com-pression did not result in significantly poorerperformance on VCV identification relative touncompressed conditions. Subjects were stillable to classify disyllabic stimuli on the basis ofcompressed envelope cues after spectral andperiodicity cues were removed. Importantly, theauthors found that compression of envelopeinformation and not the removal of periodicityinformation adversely affected subject perform-

American Academy of Audiology/Volume 14, Number 2, 2003

60

Palabras Clave: Compresión, tiempo de ataque de compresión, tiempo de liberación dela compresión, tasa de compresión, umbral de compresión, compresión del rango dinámicoamplio, fórmula de prescripción, KEMAR: Knowles Electronic Manikin for Acoustic Research,envolvente del lenguaje.

Abreviaturas: ASA2P = Alineamiento adaptativo del habla; B&K = Brüel and Kjaer; CVR= Tasa de amplitud consonante/vocal; DSL I/O = Entrada / salida del nivel deseado desensación; KEMAR = Maniquí Electrónico Knowles Para Investigación Acústica; LTASS= Espectro promedio del lenguaje a largo plazo; NALNL1 = Versión 1 del procedimientono-lineal de los Laboratorios Nacionales de Acústica; REUG = Ganancia no amplificadaen Oído Real; RMS = Raíz cuadrada media; SPL = Nivel de presión sonora; VCV = Estímulovocal-consonante-vocal; WDRC = Compresión de rango dinámica amplio.

ance for sentence recognition after spectralinformation had been removed. Van Tasell andTrine concluded that, at least for normal-hear-ing subjects, envelope cues are needed for sen-tence identification. Further evidence support-ing envelope cues in sentence recognition comesfrom Drullman (1995), who studied the role ofenvelope and fine structure cues for sentencerecognition in normal-hearing subjects. Hefound that the sentences with intact envelopecues and random fine structure cues were intel-ligible; whereas, the sentences with intact finestructure cues but random envelope cues yield-ed mean recognition scores of 17%. Overall,studies that incorporate sentences demonstratethe importance of envelope cues for sentencerecognition.

During the fitting process, hearing aidsare set to gain requirements specified by pre-scriptive formula. Three common prescriptiveformulas used to fit nonlinear instrumentsinclude Desired Sensation Level Input/Output(DSL I/O; Cornelisse et al, 1994), FIG.6 (Killionand Fikret-Pasa, 1993), and National AcousticLaboratories’ nonlinear procedure, Version 1(NAL-NL1; Byrne et al, 2001). These formulasshare the common goal of determining gainrequirements based on the level of the inputsignal. Additionally, hearing aid manufactur-ers often incorporate their own prescriptive for-mulas for fitting nonlinear hearing aids.

The practice of fitting all hearing aids to asingle prescriptive target may not be advisable.For example, Kuk and Ludvigsen (1999) claimthat no prescriptive formula accounts forattack or release times. This becomes prob-lematic because two hearing aids with differentresponse times may respond differently in real-world situations under compression even ifboth hearing aids are matched to the same pre-scriptive formula. Kuk and Ludvigsen claimthat the real-world dynamic outputs of hearingaids with similar short attack and releasetimes approximate the predicted staticinput/output curve (i.e., the response assumedby prescriptive formulas). Conversely, theyclaim that if the attack time were substantial-ly shorter than the release time, then the aver-age real-world dynamic output of the hearingaid would be below that assumed by the staticor prescription curve. They assert that match-ing prescriptive targets is acceptable with fast-acting WDRC instruments, but not for slower-acting hearing aids. Thus, Kuk and Ludvigsenquestion the value in matching prescriptivetargets in hearing aids with longer release

times.Given that compression circuits are pres-

ent in most hearing aids and that prescriptivemethods are used routinely during hearing aidfittings, the nature of their interactions shouldbe clarified. This study investigated the influ-ence that variations in specific compressionparameters produced on a dynamic speech sig-nal following adjustment of the gain of a hear-ing aid according to one of four prescriptive for-mulas designed for nonlinear hearing aids.Based on previous literature, it was expectedthat a short release time would smooth theamplitude envelope of speech relative to a longrelease time. Also, based on claims by Kuk andLudvigsen (1999), it was expected that thehearing aid would respond to a dynamic speechsignal with relatively greater output under ashort release-time than a long release-timecondition, even if the hearing aid responsematched the same target for both releasetimes. Differences across prescriptive formulaswould be expected because of inherent differ-ences in prescribed gain and compression ratio.This study addressed two questions:

(1) For a prescribed setting, do changes in compression release time differentially affect the long-term-average-speech spectrum (LTASS),the amplitude envelope, and phoneme amplitude of a three-sentence speech signal?(2) Which compression release time has the greatest effect on a speech signal for each prescriptive condition?

METHOD

Pre-recorded running speech was presentedin the soundfield and received by a

Knowles Electronics Manikin for AcousticResearch (KEMAR) in one unaided and fouraided conditions. Responses were recorded andstored for analysis offline. Repeated measureswere made of the recording conditions to assessmeasurement error. To determine the overalland long-term output of the hearing aid, theLTASS was measured in each condition.Phonemic output was measured to observegeneral phonemic trends and to corroborateLTASS results. For this, absolute level (in dBSPL) was measured for each phoneme in thespeech signal. Finally, CVRs were measured toassess the degree of alteration in the amplitudeenvelope.

Interactions of Release Time and Fitting Formula/Ellison, Harris and Muller

61

Journal of the American Academy of Audiology/Volume 14, Number 2, 2003

Speech Signal

Running speech was selected to allowmeasurement of the effect of compression on adynamic signal. The signal was pre-recordedand consisted of three sentences:

(1) Peter will keep at the tall chilly peak (pit∂^ wIl kip æ/tθ/� tɔl tʃIli pik).(2) The blue spot is on the thin key again (θ� blu spɑd Iz ɑn /ð�/ θIn ki əgεn).(3) When he comes home we’ll feed him fish(wεn /hi/ k�ms hom wIl fid /hI/m fIʃ).

These sentences were chosen for three rea-sons. First, they include most phonemes foundin American English, including all voiceless con-sonants and all vowels except diphthongs and /e,3, U, 3^, a/. An adequate variety of vowels wasneeded to measure relative changes in theiramplitude and to calculate the CVR. Voicelessconsonants were needed to exclude voicing inthe contrast between vowels and consonants tofacilitate calculation of CVRs. Second, vowelsand voiceless consonants are represented in theinitial, medial, and final position. Third, thesesentences represent running speech that a lis-tener might encounter in a real-world situation.

The sentences were spoken by a male talk-er and were recorded using a microphone(Sennheiser 825S) and a microphone amplifier(Tucker-Davis Technologies MA2) with thegain set to 35 dB. The signal was routed to asound card (Sound Fusion Wave) and wasrecorded using speech recording and analysissoftware (Praat 3.8.69) at a monaural 44100Hz sample rate.

During the recording, the talker repeatedeach sentence three times. From this sample,the middle sentence was selected and placedonto one final recording. Sentences were sepa-rated by 400 ms of silence. The recording wasconverted to a 16-bit WAV file. On the recording,a 1000 Hz marker tone of 140.7-ms duration and

19 dB below the overall level of the recordingwas placed before the onset of the first sentenceto mark the beginning of the recording.

Hearing Aid Description

An Oticon Digifocus II Power behind-the-ear digital hearing aid was used because itsrelease time can be modified. This hearing aidcontains two channels with three frequencybands in the low-frequency channel and fourfrequency bands in the high-frequency channel(cutoff frequency = 1500 Hz). Other thanrelease time, all parameters remained constantacross measurement conditions. The compres-sion release time was either 40 or 640 ms foreach measurement condition and was keptequal for both hearing aid channels. Theserelease times represent the extreme settingswithin the hearing aid and are within the rangeof release times implemented in commonly usedhearing aids currently on the market.

Prescriptive Formulas and Hearing LossConfiguration

Four nonlinear prescriptive fitting formu-las were used to set the gain of the hearing aid:DSL I/O (Cornelisse et al, 1994), FIG.6. (Killionand Fikret-Pasa, 1993), NAL-NL1 (Byrne et al,2001), and the Adaptive Speech Alignment(ASA2p) algorithm. The former three representcommon nonlinear prescriptive formulas thatrecommend fitting to the static input/outputgain curve (Kuk, 2001). Digital Oticon hearingaids incorporate the ASA2p. Oticon describesASA2p as an algorithm that compensates forrecruitment and controls for the upwardspread of masking by incorporating syllabiccompression in the low frequency channel, andattempts to achieve comfort and speech intelli-gibility by adapting the gain for changes in

62

Table 1. Prescribed Gain (in dB SPL) From 250 to 4000 Hz Using a 65 dB SPL Input For EachPrescriptive Formula and Selected Compresssion Ratio In Each Hearing Aid Channel.


63

input in the high frequency channel. Therationale is to place speech into a comfortablelistening range (Bentler and Duve, 2000) andmaintain the spectro-temporal detail of speechin the high frequencies. Because the OticonDigifocus II hearing aid is designed for theASA2p algorithm, ASA2p was included.Although a variety of instruments might be fitusing these formulae, none take into accountthe range of possible release times available. Ifprescriptive formulas attempt to approximatethe predicted static input/output curve, thenprescriptive formulas, including NAL-NL1 andDSL I/O, must assume fast release times. Incontrast, because FIG.6 was developed for usewith the K-amp circuit, it must assume vari-able release times. ASA2p incorporates timeconstants described by the algorithm.

NOAH 2.0a was employed to display pre-scriptive targets based on a hearing loss thatresembles a typical audiometric pattern of theolder adult population aged in their mid 70’s(Harford and Dodds, 1982). This configuration(see Figure 1) was used for all conditions. Table1 displays the prescribed gain between 250 and4000 Hz for a 65 dB SPL input for each fittingformula as well as the selected compressionratio in each channel that best approximatedtarget. Table 1 does not include frequenciesabove 4000 Hz because of the limited responseof the hearing aid and inability to manipulategain above this frequency.

When the hearing aid was set to the ASA2palgorithm, attack and release times werechanged to equate the settings used for thethree prescriptive formula conditions.Specifically, the attack time was adjusted from20 to 5 ms, and the release time was adjustedfrom 80 ms in the low-frequency channel and

320 ms in the high-frequency channel to 40 or640 ms in both channels. These changes didnot alter the prescribed gain for the hearinginstrument but may have altered the responseof the hearing instrument from what the man-ufacturer intended.

Hearing Aid Programming

The hearing aid was programmed toapproximate multiple input targets visually,which is a common clinical verification proce-dure. Matching to multiple input targets wasintended to account for the changing gainrequirements of nonlinear formulas at variousinput levels. OtiSet 4.20 fitting software wasused to program and manipulate the electro-acoustic parameters of the hearing aid. NOAH2.0a served as the software platform for Otiset4.20 and was utilized to observe the real-timeresponse of the hearing aid while it was pro-grammed to target. Programming the hearingaid to the ASA2p algorithm was accomplishedin Otiset by choosing the “calculate prescribedsettings” option on the Otiset toolbar withoutreal-ear verification.

Probe-microphone measures. Real-earprobe microphone measures were obtained tovisually approximate the real-ear insertiongain (REIG) of the hearing aid to prescriptivetargets. The real-ear unaided gain (REUG)was obtained so that the REIG could be deter-mined. A modulated white-noise signal gener-ated by a Madsen Aurical HiPro console andpresented through the Aurical HiPro speakerserved as the probe microphone signal used tomatch the response of the hearing aid to inputsof 40, 65, 70, and 90 dB SPL. The hearing aidgain in seven frequency bands (i.e., 250, 750,1000, 2000, 3000, 4000, and 5000 Hz) wasadjusted to adhere to the best possible visualapproximation of target gain. The process ofvisually matching hearing aid gain to targetgain verified that no visually detectable devia-tion occurred subsequent to release timemanipulations within each prescriptive condi-tion. All programming modifications per-formed in Otiset were routed to the hearing aidthrough the Aurical HiPro console and Oticonprogramming cable.

Instrumentation and Calibration

All data recordings were made from KEMARpositioned in a sound booth at a 0-degree azimuth,1

Figure 1. Audiometric configuration used to pro-gram the hearing aid across all conditions.


m from a loudspeaker, and level with the head ofKEMAR. The loudspeaker has a 6.5-in. woofer and1-in. tweeter, a response of 57 Hz to 25 kHz, acrossover of 2.2 kHz (6 dB/octave high-pass and 12dB/octave low-pass), and a sensitivity of 2.83V at 1m. The loudspeaker was powered by a 6-channelpower amplifier (Rotel, model RB-976).

Using Praat 3.9.22 software (Boersma andWeenink, 2001), a 10-s 1000 Hz calibrationtone was created to set the VU meter of theaudiometer. The level of the calibration tonewas set to the peak of the root-mean-square(RMS) envelope of the speech signal (excludingthe 400-ms silent intervals).

A Type I sound level meter (Brüel and Kjær[B&K], model 2204) and 1-in. condenser micro-phone (B&K, model 4145) were used to calibratethe signal in the sound field. The level for thecalibration tone measured at the ear of KEMARin the soundfield was 76 dB SPL. The range ofthe level of the speech signal corresponding tothis setting was 65 to 77 dB SPL, and is withinthe range of normal conversational level speechwhen the speaker’s own voice is accounted for(Skinner, 1988). Because the hearing aid com-pression kneepoint was fixed at 50 dB SPL, thislevel ensured that the hearing aid would be incompression during signal presentation.

The sound level meter and microphonewere checked for calibration at 110 dB SPL and94 dB SPL using a Quest Electronics calibrator(model CA-12B) and a B&K acoustical calibra-tor (model 4231), respectively. The real-earprobe microphone was calibrated with a 65 dBSPL swept pure tone.

Signal Delivery and Recording

The signal was presented via media software(Windows Media Player, version 6.01.05.0217) ona computer (Hewlett Packard Pentium III) androuted to the loudspeaker through an AuricalHiPro console. The signal level was controlledusing an audiometer (Madsen Aurical).

The signal was recorded via a 1/2-in.microphone (B&K, type 4192), which was cou-pled to the Zwislocki coupler of KEMAR. Aflexible adapter (model UA0122) coupled the ?-in. microphone with a preamplifier (B&K,model 2669C). The signal was then led to a sig-nal conditioner (B&K Nexus, model ZX2690)and routed into a programmable attenuator(Tucker-Davis PWS 25). The attenuator wasset to 0 dB under unaided conditions and to 24dB under aided conditions. These settings per-

mitted an optimized signal for computer pro-cessing. From the attenuator, the signal wasrouted to a sound card (Sound Fusion Wave)situated in a computer (Hewlett PackardPentium III). All signals were recorded at asampling rate of 44100 Hz and stored in WAVformat on one channel.

Reference tone. Because the level calcu-lation within the Praat software is based on anarbitrary reference, a reference tone with aknown level (94 dB SPL, B&K calibrator 4231)was recorded to convert phoneme levels fromspeech signal data recordings into absolute levels. To determine the absolute level ofphonemes within a data recording, the differ-ence between the 94 dB SPL calibration signaland the recorded level observed in Praat wasadded to the RMS calculation for each phoneme.

Estimation of measurement erroracross data recordings. To estimate themeasurement error expected from a routine fit-ting, each condition was recorded on five sepa-rate occasions.

Procedure

The audiogram was entered into NOAH2.0a. The sound level meter and microphonewere checked for calibration. The condensermicrophone of the sound level meter was hungfrom the ceiling so that it rested at the ear-levelof KEMAR. The soundfield and the VU meterof the audiometer were calibrated. The 94 dBSPL reference tone was recorded. The B&K 1/2-in. microphone was then situated in KEMAR.

The probe microphone was calibrated andplaced in the ear canal of KEMAR. Using amodulated white-noise signal, a real-ear unaid-ed response curve was obtained. The attenua-tor was set according to the signal condition. Inunaided (i.e., baseline) conditions, twosequences of the speech signal were presented,recorded, and saved. In aided conditions, thehearing aid was placed on the right ear ofKEMAR and connected to the HiPro console.The hearing aid was programmed while set toa release time of 640 ms. The speech signalwas presented, recorded, and saved as in theunaided conditions. Compression release timewas adjusted to 40 ms and the signal was againpresented, recorded, and the recording storedfor later analysis. This procedure was repeatedon five separate occasions so that each condi-tion was recorded five times.

64

Data Analysis

Data preparation. The original speechrecording was phonemically segmented into 68units. To compare the level of phonemes across alldata recordings, phonemic time segments deter-mined from the original speech signal recordingwere matched across all data recordings. This wasaccomplished by representing the onset of themarker tone (from the original speech signal andsubsequent data recordings) as the beginningpoint of all data recordings. The area following thelast phoneme was removed on all recordings toestablish a common endpoint.

MATLAB 6.1.0.450 was used to calculate RMSvalues within each phonemic time segment and theLTASS in one-third octave bands between 250 and5000 Hz for each data set. This frequency range rep-resented the frequency bands that are adjustablewithin the hearing aid.

Measures and analysis. Measures foranalysis included: LTASS in relative dB, CVR indB, and the phoneme amplitudes in dB SPL.

From the determined RMS levels, thephoneme amplitudes in each segment of the datarecordings were converted to dB SPL. The level forstops was based on burst and aspiration duration(Freyman and Nerbonne,1989;Hickson and Byrne,1997). Each condition was compared to baselineand other aided conditions as a function of release-time variation. The first phoneme from all record-ings was removed in the analysis due to minimalcompression effects. All data sets were equivalentlybalanced, resulting in 32 total phonemic units.

The CVR was determined in dB by subtract-ing the consonant from the vowel dB SPL to mini-mize negative values. Only voiceless consonant-vowel contrasts were used for these calculationsbecause they represent the greatest contrasts amongspeech elements. An average CVR was determinedin each condition and used for comparisons betweenconditions and to baseline as a function of release-time variations. The first consonant-vowel contrastin all data recordings was omitted due to minimalcompression effects. All data sets were equivalentlybalanced, resulting in 18 consonant-vowel contrasts.

The LTASS is expressed in relative dB andrepresents the energy in 14 one-third octave bands.A common reference of 0.00000000001 was used toconvert RMS to dB values. This reference wasused to avoid results less than 1.0 and hence, neg-ative dB values. Relative dB results were derivedand used for comparisons between conditions andto baseline.

Statistical analysis. A multiple comparisonANOVA was performed for all measures. The

influence of four variables (i.e., frequency, prescrip-tive formula, release time, and trial) and theirinteractions upon the LTASS were assessed. Also,four variables (i.e., phoneme, prescriptive formula,release time, and trial) and their interactions uponthe CVR and phoneme level were assessed. Analpha of 0.02 was used to adjust for multiple compar-isons. A Tukey Studentized-Range (HSD) post-hocanalysis (α=0.02) was performed for all measures.

Generalizability theory was applied to allmeasures to predict the proportion of expectedvariance of each of the independent componentvariables discussed for ANOVA upon their respec-tive dependent variables.

For descriptive, ANOVA, and Generalizabilitytheory analysis, CVR and phoneme amplitudedata sets were balanced by randomly removingrepeated observations and applying changes uni-formly across trial recordings. This created datasets with homogeneity of variance and alloweddirect comparison between all statistical analysesto be made.

RESULTS

Long Term Average Speech Spectrum(LTASS)

The effect of changes in release time on theLTASS was determined by calculating the differ-ence in dB between each one-third-octave bandmeasured in the 40- and 640-ms release-time con-ditions. Figure 2 displays these mean differenceswith the hearing aid set to the ASA2p, DSL I/O,FIG.6, and NAL-NL1 algorithms, respectively.Relative to the 640-ms conditions, the mean levelsfor all of the 40-ms conditions were greater:ASA2p= 4.32 dB (sd = 0.17),DSL I/O = 3.21 dB (sd = 0.56),FIG.6 = 5.47 dB (sd = 1.05),and NAL-NL1 = 2.8 dB(sd = 0.61). FIG.6 was the most affected bychanges in release time; whereas, NAL-NL1 wasthe least affected. In all prescriptive conditions,theone-third octave bands with center frequencies of1600, 1250, and 2000 Hz were most altered withchanges in release time. The 250, 315, and 630 Hzbands were least affected by changes in releasetime under all formula conditions, except for DSLI/O in which the 5000 Hz band was the least sen-sitive; however, this applied to only one frequency.Otherwise, findings were compatible with the otherthree formulae.

The effect of each release time on the unaid-ed LTASS in each formula was determined bycalculating the difference (dB) in the LTASSbetween aided and unaided conditions (i.e.,gain). Because the hearing aid used a cutoff fre-quency of 1500 Hz, the mean gain from one-


65


66

third octave bands was assessed within a high-and low-frequency channel using a cutoff fre-quency of 1500 Hz. Results suggest that withineach prescriptive condition, the hearing aid pro-vided more gain when it was set to a 40- ratherthan a 640-ms release time. With the exceptionof measures within DSL I/O, the aided LTASS inthe low-frequency channel was less than theunaided LTASS under all conditions (i.e., nega-tive gain). This was likely the result of a combi-nation of factors. First, minimal gain was pre-scribed in the low-frequency channel; second,the signal would have activated the hearingaid’s compression in the low-frequency channeland thereby reduced the gain in that channel;third, the occluding ear mold attenuated the sig-nal reaching the ear canal of KEMAR. Thegreatest mean gain across all 14 one-thirdoctave bands occurred when the hearing aid wasset to DSL I/O with a 40-ms release time (seeFigure 3). Because of the negative low-frequen-cy gain, values in Figure 3 are below prescribedvalues. The least mean gain occurred when thehearing aid was set to ASA2p with a 640-ms

release time. Due to the degree of negative gainbelow 1500 Hz, the hearing aid provided nega-tive overall gain (mean across all octave bands)when it was set to ASA2p even though the gainin the high-frequency channel was still positive.

The results of the ANOVA for LTASS meas-ures indicated that formula, release time, andfrequency significantly contributed to the vari-ance of LTASS measures. The interactions offormula and frequency, formula and releasetime, and release time and frequency signifi-cantly contributed to the variance. Tukey HSDpost-hoc tests (α=0.02) revealed significanteffects for all formula and release-time combina-tions except for the comparison between NAL-NL1 and FIG.6 with 640-ms release times.Separate trials did not significantly contributeto the variance of LTASS measures. Tukey HSDpost-hoc tests (α=0.02) also revealed significanteffects for all formula and release-time combina-tions when compared to unaided conditions.

The results of the Generalizability theoryon LTASS measures indicated that frequency,prescriptive formula, and the interaction

Figure 2. Mean and standard deviation of LTASS differences between 40-ms and 640-ms release-time conditions within ASA2p, DSL I/O, FIG6, and NAL-NL1. The y-axis represents the difference between 40-msand 640-ms conditions (40 640 ms).

between frequency and formula contributed toroughly 97% of the predicted variance estimatesof LTASS measures. Release time contributedto less than 2% of the variance.

Consonant-Vowel Ratio

The effect of changes in release time on theamplitude envelope was determined by obtain-ing the difference (dB) in the CVR between 40-and 640-ms release-time conditions. Figure 4displays the mean differences with the hearingaid set to the ASA2p, DSL I/O, FIG.6, and NAL-NL1 algorithms. The mean CVRs for the 40-ms

conditions are less than those of the 640-ms con-ditions by 0.92 (sd=0.44), 1.51 (sd=0.41), 1.82(sd=0.62), and 1.12 (sd=0.35) dB for ASA2p, DSLI/O, FIG.6, and NAL-NL1 conditions, respective-ly. Results suggest that the CVR is most affect-ed by release-time changes with the hearing aidset to FIG.6. Within all formulas, /θ�/ was theleast affected by release-time changes.Conversely, /it/ was most affected by release-timechanges within all formula conditions.Generally, contrasts that included voiceless stopswere most sensitive to release-time changes.

The effect of each release time on the natu-ral amplitude envelope within each prescriptive


67

Table 2. The Five Phonemes Least and Most Affected by Changes in Release Time.

Figure 3. Mean and standard deviation of the differences in the LTASS between each formula/ release-timecombination and unaided condition (i.e., gain) for all one-third octave bands combined. Differences were cal-culated by subtracting the unaided LTASS from either the 40- or 640-ms conditions. Baseline (i.e., 0) repre-sents the unaided LTASS. Bars extending from baseline represent the standard deviation of the unaided con-ditions.

formula was determined by calculating the dif-ference (dB) in the CVR between aided andunaided conditions. Figure 4 displays thesemean differences with the hearing aid set toASA2p, DSL I/O, FIG.6, and NAL-NL1. Relativeto unaided conditions, negative dB values repre-sent reduced differences in level between the voice-less consonant and neighboring vowel under aidedconditions; whereas, positive values representincreased differences. These results suggest lessalteration of the natural CVR with the hearing aidset to a 640-ms release time rather than a 40-msrelease time. The greatest alteration of the naturalCVR occurred when the hearing aid was set to FIG.6and a 40-ms release time. Conversely,the least alter-ation of the natural CVR occurred with the hearingaid set to DSL I/O and a 640-ms release time.

The results of the ANOVA indicated thatformula, release time, consonant-vowel contrast,and trial significantly contributed to the vari-ance of the CVR. Also, the interactions of formu-la and release time, formula and consonant-vowel contrasts, release time and consonant-vowel contrasts, and the interactions between allthree (i.e., formula, release time, and consonant-vowel contrasts) contributed to the variance ofthe dependent variable. Tukey HSD post-hoctests (α=0.02) revealed significant effects for allformula and release-time combinations exceptfor the comparison between NAL-NL1 and FIG.6with 640-ms release times, and between NAL-NL1 and DSL I/O with 40-ms release times. Allrelease-time effects were significant. Tukey HSDpost-hoc tests (a=0.02) also revealed significanteffects for all formula and release-time combina-tions when compared to baseline.

The results of the Generalizability theory onCVR measures indicated that the voiceless conso-nant-vowel contrasts contributed to the majorityof predicted variance estimates of CVR measures.Release time alone contributed to approximately2% of the estimated variance. The combinedinteractions between the contrasts and releasetimes, and the contrasts and formulas con-tributed to approximately 10% of the predictedvariance estimates.

Phoneme Amplitude

The effect of changes in release time onphoneme amplitude was determined by calculat-ing the difference (in dB) in phoneme amplitudesbetween the 40- and 640-ms conditions. Relativeto the 640-ms conditions, the mean output for the40-ms conditions was greater by 3.12 (sd = 0.4),2.57 (sd=0.55), 4.32 (sd=0.77), 2.51 (sd=0.43) dBfor ASA2p, DSL I/O, FIG.6, and NAL-NL1,respectively. Thus for the phoneme amplitudes,FIG.6 was the most affected by release-timechanges whereas NAL-NL1 was least affected.This pattern confirms the LTASS results. Table2 displays the five phonemes least and mostaffected by changes in release time across all pre-scriptive conditions. With some exceptions, stopswere generally more sensitive to release-timechanges than were fricatives. Under all condi-tions, /o/ was the vowel most affected by release-time changes. The vowel least affected by releasetime was /I/. Under ASA2p, FIG.6, and NAL-NL1, /b/ and /w/ constituted the only units withgreater output in the 640-ms conditions than inthe 40-ms conditions.


68

Figure 4. Mean and standard deviation of differ-ences in the CVR for each formula / release-timecombination and the unaided condition.Differences were calculated by subtracting theunaided CVR from either the 40- or 640-ms condi-tions. Baseline (i.e., 0) represents the unaidedCVR. Bars extending from baseline represent thestandard deviation of the unaided conditions.

Figure 5. Mean and standard deviation of overalldifferences in the phoneme amplitude betweeneach formula / release-time combination andunaided condition (i.e., gain). Differences werecalculated by subtracting the unaided phonemeamplitude from either the 40- or 640-ms conditions.Baseline (i.e., 0) represents the unaided phonemeamplitude. Bars extending from baseline representthe standard deviation of the unaided conditions.

The effect of release time on the naturalphoneme amplitude within each formula wasdetermined by calculating the difference in dB(i.e., gain) between aided and unaided conditions.Figure 5 displays the mean differences with thehearing aid set to the ASA2p, DSL I/O, FIG.6,and NAL-NL1 algorithms. These results suggestthat within each respective formula, the hearingaid provided more gain when the hearing aidwas set to a 40- rather than a 640-ms releasetime. The most gain occurred with the hearingaid set to DSL I/O with a 40-ms release time.Conversely, the least gain occurred with thehearing aid set to ASA2p with a 640-ms releasetime. This pattern generally agrees with LTASSresults.

The results of the ANOVA indicated that theformula, release time, phoneme, and trial signifi-cantly contributed to the variance of the phonemeamplitude. Also, the interactions of formula andrelease time, formula and phoneme, release timeand phoneme, and the interactions between allthree (i.e., formula, release time, and consonant-vowel contrasts) contributed to the variance of thedependent variable. Tukey HSD post-hoc tests(α=0.02) revealed significant effects for all formu-la and release-time combinations except for thecomparison between NAL-NL1 with a 40-msrelease time and FIG.6 with a 640-ms releasetime. All release-time effects were significant.Tukey HSD post-hoc tests (α=0.02) also revealedsignificant effects for all formula and release-timecombinations when compared to baseline.

The results of the Generalizability theory onphoneme amplitude indicated that thephonemes, then formulas contributed the most tovariance estimates, and together accounted forroughly 86% of the variance. Release time con-tributed to approximately 7.5% of the estimatedvariance. The interaction between phoneme andrelease time accounted for less than 4% of theestimated variance.

SUMMARY AND DISCUSSION

The purpose of this investigation was to studythe interactive effects of hearing aid release

time and prescribed gain on the output of a hear-ing aid in response to a speech signal. Resultssupported that within ASA2p, DSL I/O, FIG.6, andNAL-NL1, changes in release time had small butsignificant effects on LTASS, CVR, and phonemeamplitude. Overall, gain and output were greaterwith the hearing aid set to a 40-ms than a 640-msrelease time. High frequencies were more affect-ed by changes in release time than were other fre-

quencies. Under 40-ms release time conditions,differences in level between the vowels and neigh-boring voiceless consonants were reduced.Phoneme amplitudes were greater with the hear-ing aid set to a short release time, which support-ed LTASS results. Stops were generally moreaffected by changes in release time than werefricatives. Across all measures, the effect ofchanges in release time was greatest for the FIG.6prescription. These results support those of otherinvestigators and provide the basis for some limit-ed clinical implications.

Both Van Tasell (1993) and Kuk (1998)reported that long release times preserve theamplitude fluctuations in the speech envelope;whereas, short release times smooth the enve-lope. Our results confirmed this effect.Frequencies above 1500 Hz were more affectedby changes in release time relative to low fre-quencies. Several authors (Drullman, 1995; VanTasell and Trine, 1996; Souza and Kitch, 2001)have demonstrated the importance of envelopecues for sentence recognition. A positive correla-tion exists between the amount of access tospeech envelope cues and performance on wordor syllable tasks for subjects with profound hear-ing loss (Erber, 1972; Boothroyd et al., 1988,).Kuk and Ludvigsen (1999) asserted that if theattack time were substantially shorter than therelease time, then gain prescribed by targetwould be insufficient. Because most nonlinearprescriptive formulas (with the exception ofFIG.6) assume short release times, fitting long-release-time hearing aids to prescription mayprovide less gain and output than what is need-ed by the listener in the real world.

Among the prescriptive formulas used in thecurrent study, FIG.6 prescribed the greatest com-pression ratio and had the most effect on thedependent variables with release time changes.This result is not unexpected: the greater the com-pression ratio, the larger the recovery in gain aftercompression. The effect would also likely begreater with more prescribed gain. For example, anonlinear hearing aid programmed for a profoundhearing loss would utilize higher compressionratios and more gain, and perform within closerproximity to its upper limit. This would likelyresult in greater gain, output, and speech envelopediscrepancies with changes in release time.Therefore, failure to account for release time couldbe detrimental for speech intelligibility in individ-uals with profound hearing loss because of theirgreater reliance on envelope cues (Erber, 1972;Boothroyd et al, 1988) and the need for high levelsof gain. The two-channel hearing aid used in the

69


study would have likely altered the speech enve-lope more than a single-channel device (Plomp,1988). Such release-time effects, therefore, may beless consequential in a single-channel device.

It is unknown how the acoustic changesmeasured in this study would affect a listener’sperformance. If variations up to 8 dB (e.g., FIG.6at 2000 Hz) in the higher frequencies are clini-cally significant for speech intelligibility, thenchanging the release time or fitting a slow-release-time hearing aid to prescription couldhave adverse effects on speech intelligibility.Despite the statistically significant effect thatchanges in release time had on the dependentvariables, such changes resulted in smallabsolute differences between the 40- and 640-msconditions. Such small differences might haveminimal clinical consequences. Release timeaccounted for a small proportion of the estimatedvariance relative to other independent variables(e.g., frequency, vowel-consonant contrasts, andphoneme). Therefore, despite the significanteffects from changes in release time on speech, cli-nicians may need to be relatively more concernedwith other variables. However, it is possible thatif release time were ignored, then the practice offitting hearing aids to prescriptive targets mightresult in poorer performance for a hearing-impaired listener. Because the practice of match-ing hearing aids to target is essential and gener-ally more relied upon in the beginning of the fit-ting process, hearing aid acceptance could beaffected, especially for first time hearing aidusers. Direct evaluation of these effects would beneeded to make such a determination.

There were limitations to the study thatshould be considered when interpreting the find-ings. Using averaged data, such as the LTASS,may undervalue the effect of release time on thelowest amplitude elements of speech. Also, ourinstrument maintained a fixed release time whencompression ratio was altered. To achieve this,the slope of the release changed: the higher thecompression ratio, the steeper the slope when therelease time was held constant. This interactionbetween release time and compression ratiomight have had an effect that we were unable tomeasure or to account for in the analyses.

At present, the degree of change in releasetime needed to affect intelligibility significantlyis not fully understood. The determination ofsuch effects through future research could assistclinicians in achieving more successful hearingaid fittings. It is important to keep in mind, how-ever, that the practice of setting a hearing aid toa prescriptive formula is a verification procedure(Kuk, 2001) and does not necessarily reflect the

final fitting. The effects of release-time changesfor a hearing aid programmed for more severehearing losses and the determination of correc-tion factors when setting slow-release-time hear-ing aids to nonlinear prescriptive targets areboth areas that deserve further investigation.

REFERENCES

Balakrishnan U, Freyman RL, Chiang Y, et al. (1996).Consonant recognition for spectrally degraded speech as afunction of consonant-vowel intensity ratio. J Acoust Soc Am99:3758-3769.

Bentler RA, Duve MR. (2000). Comparison of hearing aidsover the 20th century. Ear Hear 21:625-639.

Boersma P, Weenink D. (2001). Praat 3.9.22: A system fordoing phonetics by computer [Computer software].Amsterdam: Institute of Phonetic Sciences.

Boothroyd A, Springer N, Smith L, Schulman J. (1988).Amplitude compression and profound hearing loss. J SpeechHear Res 31:362-376.

Byrne D, Dillion H, Ching T, et al. (2001). NAL-NL1 procedurefor fitting nonlinear hearing aids: Characteristics and com-parisons with other procedures. J Am Acad Audiol 12:37-51.

Cornelisse LE, Seewald RC, Jamieson DG. (1994). Widedynamic range compression hearing aids: The DSL [i/o]approach. Hear J 47(10):23-29.

Drullman R. (1995). Temporal envelope and fine structurecues for speech intelligibility. J Acoust Soc Am 97:585-592.

Erber NP. (1972). Speech-envelope cues as an acoustic aid tolipreading for profoundly deaf children. J Acoust Soc Am51:1224-1227.

Freyman RL, Nerbonne GP. (1989). The importance of conso-nant-vowel intensity ratio in the intelligibility of voicelessconsonants. J Speech Hear Res 32:524-535.Gordon-Salant S. (1986). Recognition of natural andtime/intensity altered CVs by young and elderly subjectswith normal hearing. J Acoust Soc Am 80:1599-1607.

Gordon-Salant S. (1987). Effects of acoustic modification onconsonant recognition by elderly hearing-impaired subjects.J Acoust Soc Am 81:1199-1202.

Harford ER, Dodds E. (1982). Hearing status of ambulatorysenior citizens. Ear Hear 3:105-109.

Hickson L, Byrne D. (1997). Consonant perception in quiet:Effect of increasing the consonant-vowel ratio with compres-sion amplification. J Amer Acad Audiol 8:322-332.

Killion MC, Fikret-Pasa S. (1993). The 3 types of sen-sorineural hearing loss: Loudness and intelligibility consid-erations. Hear J 46(11):31-36.

Kuk FK. (1998). Rationale and requirements for a slow-act-ing compression hearing aid. Hear J 51(6):45-53, 79.

Kuk FK. (2001, April). Verification issues with nonlinearhearing aids. Paper presented at the annual convention ofthe American Academy of Audiology, San Diego, CA.

Kuk FK, Ludvigsen C. (1999). Variables affecting the use ofprescriptive formulae to fit modern nonlinear hearing aids. JAmer Acad Audiol 10:458-465.


70

Plomp R. (1988). The negative effect of amplitude compres-sion in multichannel hearing aids in light of the modulation-transfer function. J Acoust Soc Am 83:2322-2327.

Preves DA, Fortune TW, Woodruff B, Newton J. (1991).Strategies for enhancing the consonant to vowel intensity ratiowith in the ear hearing aids. Ear Hear 12(Suppl. 6):139S-153S.

Rosen S. (1992). Temporal information in speech: Acoustic,auditory and linguistic aspects. Philosophical Transactionsof the Royal Society of London. Series B: Biological Sciences,336, 367-373.

Sammeth CA, Dorman MF, Stearns CJ. (1999). The role ofconsonant-vowel amplitude ratio in the recognition of voice-less stop consonants by listeners with hearing impairment. JSpeech Lang Hear Res 42:42-55.

Skinner MW. (1988). Hearing aid evaluation. Englewood,Cliffs, NJ: Prentice Hall.

Smith LZ, Levitt H. (1999). Consonant enhancement effectson speech recognition of hearing-impaired children. J AmAcad Audiol 10:411-421.

Souza PE. (2000). Older listeners’ use of temporal cuesaltered by compression amplification. J Speech Lang HearRes 43:661-674.

Souza PE, Kitch V. (2001). The contribution of amplitudeenvelope cues to sentence identification in young and agedlisteners. Ear Hear 22:112-119.

Souza PE, Turner CW. (1996). Effect of single-channel com-pression on temporal speech information. J Speech Hear Res39:901-911.

Van Tasell DJ. (1993). Hearing loss, speech, and hearing aids.J Speech Hear Res 36:228-244.

Van Tasell DJ, Trine TD. (1996). Effects of single-band syl-labic amplitude compression on temporal speech informationin nonsense syllables and in sentences. J Speech Hear Res39:912-922.


71

Documents

Interactions of Hearing Aid Compression Release Time and Fitting