Perceived soundquality of reproductions with different frequency responsesand sound levels AIf Gabrielsson Departmentof TechnicalAudiology, Karolinska Institute, Stockholm, and Department of Psychology, UppsalaUniversity, Uppsala, Sweden BjOrn Hagerman,Tommy Bech-Kristensen,and GOran Lundberg Departmentof TechnicalA udiology,Karolinska Institute, K TH, S-10044 Stockholm,Sweden ( Received27 February1989; accepted for publication6 April 1990) Threeprograms (female voice, jazz music,and pink noise)were reproduced using four differentfrequency responses and two differentsound levels. Fourteen normal hearing subjects listenedto the reproductionsvia carphonesand judged the soundquality on sevenperceptual scales(loudness, clarity, fullness,spaciousness, brightness, softness/gentleness, and nearness} and a fidelity scale.Significant differences among the reproductionsappeared in all scalesand couldbe attributedto the differencesin frequencyresponse or soundlevel or both. Interactions betweenthe reproductionsand the programscould be explainedby the relationsbetween the spectrumof the programsand the frequencyresponses used. The resultsfor the noiseprogram weresimilar to thosefor thejazz musicprogram. PACS numbers:43.66.Lj, 43.88.Md [NFV] INTRODUCTION crease(dullness and softness/gentlenessdecrease) with ris- ing frequencyresponse toward higher frequenciesand/or The perceivedsound quality of sound-reproducingsys- falling responsetoward lower frequencies(cf. alsoStevens tems-such as loudspeakers,headphones, and hearing and Davis, 1938,pp. 163-166, for densityand brightnessin aids--is multidimensional;that is, it is associatedwith a sinusoidaltones; Bismarck, 1974). Fullnessis favoredby a numberof perceptualdimensions. By meansof multivariate broadfrequency range and relatively more emphasis on low- methodsused in experimentalpsychology, we wereable to er frequencies(cf. Stevensand Davis, 1938,p. 161,for vol- showthat the perceivedsound quality can be describedin ume in sinusoidaltones). Clarity, spaciousness,and (to termsof dimensionssuch as clarity, fullness, brightness ver- someextent) nearnessare likewisefavored by a broad fre- susdullness, sharpness versus softness/gentleness, loudness, quencyrange, often with a certainemphasis on midhighto spaciousness,nearness, and absenceof extraneoussounds highfrequencies. The effectsof extraneoussounds, e.g., hiss, (Gabrielsson,1979a; Gabrielsson and Sjfigren, 1979a). Rat- may be relievedby reducedresponse at high frequencies. ing scalesfor thesedimensions have beensuccessfully used Generally,the resultsalso depend on the characteristicsof to provideperceptual descriptions of loudspeakersand other the (musicor speech)programs that are reproduced.There sound-reproducingsystems (Gabrielsson and Lindstr/Jm, are thusinteractions between the reproductionsystems and 1985; Gabrielsson,1987; Gabrielssonet al., 1988). Overall the programs. evaluationsof thesystems in termsof fidelityor pleasantness Another importantphysical factor is the soundlevel. may be regardedas weightedcombinations of the separate The availableevidence (e.g., Gabrielssonand Sj/Sgren, perceptualdimensions. The weight given to each dimen- 1979a) indicatesthat an increasein soundlevel will usually sion-that is, how important it is for the overall evalua- increasethe perceivedfullness, spaciousness, and nearness tionsdependson the characterof the programto be repro- aswell as sharpness and brightness; a decrease in soundlevel ducedand the listener'searlier experiences. givesthe oppositeresults. Increased sound level may also The relationsbetween the perceptualdimensions and contributeto increasedclarity, althoughonly up to a certain variousphysical properties of the systemsare complexand level at which overloadingin the auditory systemmay occur still largely unknown.Among the physicalvariables the fre- (Kryter, 1970, p. 52). There may be interactionsbetween quencyresponse is often consideredas the mostimportant. the soundlevel, on the one hand, and the frequencyresponse Its effectson the perceptualdimensions have been explored and/or the spectrumof the programon the other hand.For in a posthoc manner by studyingthe frequencyresponse of instance,a programreproduced by a systemwith boosted the systemsreceiving different ratings in the respectivedi- treblemay soundeven sharper and brighter if the soundlevel mensionsand alsomore directly by experimentalmanipula- is increased,while a reproductionwith boostedbass will tion of the frequencyresponse (Gabrielsson et al., 1974;Ga- probablysound even duller if the soundlevel is raised. brielssonand Sj/Sgren,1979a,b; Gabrielsson et al., 1986, Becauseof suchcomplex interactions and alsobecause 1987, 1988;cf. alsosimilar approachesby Komamura eta/., of theposthoc character of certainresults referred to above,a 1977; K/Stter, 1968; Staffeldt, 1974; Toole, 1986; Toole and further experimentwith systematicmanipulation of the fre- Olive, 1988). The resultsof our investigationsindicate that quencyresponse and the soundlevel was conducted.The the frequencyresponse can affect any of the above-men- purposewas to investigatethe effectson the perceptualdi- tionedperceptual dimensions. Brightness and sharpnessin- mensionsof four markedly differentfrequency responses 1359 J. Acoust.Soc. Am. 88 (3), September1990 0001-4966/90/091359-08500.80 @ 1990 AcousticalSociety of America 1359 (flat, boostedat low, midhigh,and high frequencies)and before binaural (diotic) presentationto the listeners two differentsound levels in the reproductionof threepro- through Sony Walkman MDR-E262 earphones.The fre- gramsincluding speech, music, and pink noise. quencyresponse of the earphonesis shown in Fig. 2. The steepcutoff at about6 kHz isdue to the antialiasinglow-pass I. METHOD filter described below. A. Programs The filterswere implemented as digital FIR filtersusing a generalpurpose measuring device (TAMP3) developedin Three programswere used:(1) pink noise( -- 3 dB/ our department.It can be usedfor measuringthe frequency oct), monophonicrecording; (2) femalevoice readinga responseof linear,time-invariant systems and also for digital fairy-talein an anechoicchamber, monophonic recording; filteringof signalsin real time. It is equippedwith antialias- and (3) jazz music, excerpt from "Ole Miss" by W. C. ing filtersand fast 12-bit A/D and D/A converters.Various Handy, performedby The Peoria Jazz Band in an auditor- typesof input amplifiers,output amplifiers, and attenuators ium. Phonographrecord: Opus 3, 79-00, Testskiva1: Per- canbe connected to the processor.All unitsare controlled by spektiv.The excerptwas copieddirectly from the stereo- a personalcomputer. phonicrecording on the mastertape, but was played The samplingfrequency of the digital filter had to be monophonically.to the listener. restrictedto 20 kHz, and an antialiasinglow-pass filter was Each programlasted for about 1 min. The pink noise setto 6.7 kHz. Four differentfilters were implemented. One was chosen to serve as a neutral reference. The female voice filter had a flat response,that is, no filteringat all. The other in the anechoicchamber has most of its energybelow 1 kHz, three filters gave about 9 dB amplificationbelow 200 Hz, especiallybetween about 130 and 700 Hz, while the jazz around 1 kHz, and around4 kHz (seeFig. 3). (The lowest music has a considerablybroader frequencyrange With a filter couldnot be madesymmetrical because of certainlimi- boostaround 100 Hz (seeFig. 1). The programswere low- tationsin the equipment.Below 100 Hz thereis little energy passedat 6.7 kHz for reasonsexplained below. due to the cutoffof the earphones;cf. Fig. 2.) Thesefilters will be referredto as the L (for low), M (midhigh), and H B. Reproduction system (high) filter, respectively. A tape recorder (Telefunken Magnetophon28) was The soundlevels were set to representan approximately usedto reproducethe programs,which were then filtered natural level of the respectiveprogram when listenedto in the earphonesthrough the filter with fiat response.Mea- suredby a coupleraccording to IEC 711 fitted into the KE- MAR manikin, the A-weightedsound level for the pink noisewith the fiat responsewas about 68 dB, for the female { 10dB voiceabout 56 dB, and for thejazz musicabout 80 dB. For comparisoneach program was also presentedat a l0 dB lower level. The filtersthemselves caused certain changes in the o,zo 0.5(] 2.oo 5.o({ 0.1 1.o soundlevel. These effects were different for differentpro- kH• gramsdepending on their spectrum(Fig. l ). For the pink noisethere was practically no differencein the A-weighted soundlevel betweenthe fiat responseand the L filter, while the M filter gave an increaseof 2 dB, and the H filter an voice increaseof 6 dB. For thejazz musicthere was again no differ- i 10dB encein the A-weightedsound level between the fiat response and the L filter, while the M filter gavean increaseof about$ dB and the H filter an increase of about 3 dB. For the female O.ZO 0.50 Z.00 voicethe L filter increasedthe A-weightedsound level by 0.1 1.0 kXz i 10d8 I10dB 0.20 0.50 2,00 5.OO 0.1 l.O .01 .05 .1 .5 I 5 10 kHz kHz FIG. 1. Long-timeaverage spectrum (LTAS) for the threeprograms. LTAS wascalculated from the autocorrelation function of theprogram and FIG. 2. Frequencyresponse of the earphonemeasured on a manikin (KE- smoothed across octaves. MAR) equippedwith an earsimulator according to IEC 711. 1360 J. Acoust.Soc. Am., Vol. 88, No. 3, September1990 Gabrielssonat a/.: Perceivedreproduction quality 1360 o 2 3 q 5 G ? 8 9 IO I 5d8 • ......... • ......... • ......... • ........