Assessment of Audio Sound Quality Based on Psychoacoustics

Transcription

1 Proceeding of th International Congre on Acoutic, ICA -7 Augut, Sydney, Autralia Aement of Audio Sound Quality Baed on Pychoacoutic Yu Ting Tai and Jin H. Huang Electroacoutic Graduate program, Feng Chia Univerity, Taichung 7, Taiwan (ROC) PACS:.7.Zz,.7.Yy,.7.St,.7.Cd ABSTRACT Sound quality i an important iue in ound product today, covering a range of field form muic performance to mechanical noie, and i related to human aural repone. Many meaurement aement item for ound quality have been defined including frequency and loudne. Muic alo include all ubjective characteritic of ound. Sound allow people to appreciate their urrounding through auditory organ, and litener naturally anticipate enjoyment of muic. In brief, timbre i determined by hearing enation and atifaction i determined by both ound imaging and artitic content. When muic i played, litener pay attention to hearing enation firt and atifaction econd. But timbre of feeling i difficult to expre objectively a litener ubjective feeling cannot be accurately meaured by acoutic meaurement equipment. To combine objective analyi and pychoacoutic to reinterpret the ratio of timbre of feeling in ound quality, thi paper preent an aement model for the ound quality of audio performance baed on pychoacoutic theory. The model incorporate auditory roughne and pecific loudne that are deemed the caue of the quality of audio performance. From the model, the optimum curve for auditory roughne i preented. Furthermore, the hearing balance of high audio fidelity and hearing atifaction rank are propoed. Experimental reult how that the model can be applied not only to meaure the ound quality of audio ignal but alo to ae ound quality qualitative comparion of high fidelity loudpeaker. The reult alo demontrate that the propoed aement model i capable of expreing ubjective ound quality uccefully. INTRODUCTION Sound quality i a vutal iue for every audio product that can be either ubjectively or objectively. Subjective aement meaure human auditory perception directly. Given that hearing i not a purely mechanical phenomenon of wave tranmiion and it alo involve both perception and cognitive procee in a peron litening experience, ubjective experiment are therefore eential even though thee experiment are in general inefficient, time conuming and context dependent. On the other hand, objective method require through invetigation of the relation between the meaurement and perceived quality. Theoretically thi can be achieved by developing all tranformation function of human auditory ytem with the knowledge of human anatomy. it i deirable that ubjective judgement or aement can be replaced, or at leat complemented, by an objective meaurement method. Previou tudie on loudpeaker analyi, including thoe by Leong [] and Hirahara [], ued frequency repone meaured to analyze peaker quality by machine. Kallinen [] ued frequency repone meaured by machine in combination with ubjective hearing tet to dicu ound quality. Sung [] developed an objective meaurement indicator to dicu people perception of timbre. Voinle and Briolle [] ubjectively dicued in ound quality in term of mathematical tranfer function. In contrat the above-mentioned tudie, Fatl [6] developed a pychoacoutic theory uing ear perception and pychoacoutic that to better expre ound quality change. Meaurement data and perception weighted curve of ear experience can be ued to expre ome ubjective characteritic. Previou tudie have not accurately defined the threhold of people perception of ound quality. Thu, the aim of thi tudy i to develop a model for determining uch threhold. A pychoacoutic can expre ear reaction, they are often ued to meaure noie. Chatterley [7] compared pychoacoutic meaure of harpne, roughne, tonality and loudne ued in ubjective and objective teting of mechanical noie. Although pychoacoutic i often dicued in iue of noie, thi doe not mean that pychoacoutic cannot be applied to muic. In audio compreion, the mot important application i Pychoacoutic-model [8, 9], uing making effect [] to identify ear enation and then to execute audio compreion. Due to the ucce of MPEG coding, good and bad apect of compreion can be ued to provide information on the objective indice of ear timbre a well a quality threhold in ubjective timbre. Therefore, thi paper etablihed a Sound Quality Aement Model to determine ound quality variation equilibrium to dicu the index threhold of ound quality. Thi model olve the problem of uing hearing enation parameter in the identification of ubjective ound quality. Empirical method included known compreion ignal pectrum analyi of ound quality and verification of the effectivene of our method. In addition, headphone were ued a a meaurement material to compare ound quality by muic playing. ICA

2 -7 Augut, Sydney, Autralia Proceeding of th International Congre on Acoutic, ICA SOUND QUALITY ASSESSMENT MODEL The ound quality aement model ue ubjective quantization method of pychoacoutic including Auditory Roughne and Specific Loudne to oberve and analyze the change in ound pectrum. Spectrum data operation of Auditory Roughne i carried out according to the Roughne calculation method of Vailaki [, ]. The Roughne data analyi of Vailaki ue time to expre Roughne total value of each frequency. However, due to fluctuation in ignal time, it i difficult to explain Roughne value of each frequency on hearing enation. Therefore, in thi paper, we divided unequal frequency filter into hearing perception band according to the method of []. Uing Vailaki method to calculate Roughne amount, Roughne amount from hearing perception band in each frequency can be clearly expreed, and thi can be referred to a Auditory Roughne. Specific Loudne calculation i carried out according to the method of Zwicker [], uing ubjective method to repreent volume loudne of each frequency. If variation in loudne between high and low i mall, ound balance i good. A volume loudne cannot expre ignal in timbre fluctuation of each frequency, it mut be coordinated with Auditory Roughne decribed in the previou paragraph. What proportion of Specific Loudne [-6] and Auditory Roughne can expre better ubjective ound? Thi ratio of Hearing Balance can expre hearing balance to dicu ound performance. The importance of the Sound Quality Aement Model i that the data on Hearing Balance can be ued to develop ummation weighted Satifaction Rank to dicu threhold of ound quality. kf where k and f N i the ampling rate in Hz. T The STFT bin number i k. Each bin X( ) of the STFT can be regarded a a ample of the complex ignal at the output of a lowpa filter with input X ( n) W ( n m) e j kn ; thi ignal i Xn ( ) frequency-hifted o that frequency w k i moved to Hz. In thi interpretation, the hop ize R i the downampling factor applied to each bandpa output, and the analyi window Wn ( ) i the impule repone of the anti-aliaing filter ued with the downampling. However, X( ) i a function of two variable: time and frequency. After obtaining time and frequency, firt, cancel the weak ignal below db, and then elect each time with frequency to determine the filter bank a critical-band rate a in the following equation. f f Bn *arctan.76.*arctan khz 7. khz where B n i the index number in the frequency cale. The individual critical band have bandwidth f b a follow. [] Auditory Roughne Analyi Model Figure how a flowchart of auditory roughne analyi model. Computation of the auditory roughne analyi model conit of the following tep: f fb 7. khz.69 [] X(n) STFT Separate Frequency Band into B (k) B (k) RAM RAM Rt Eq. and Eq. decribe the dependence of critical-band rate alo called []. It i important to implement the filter deign. The bark computation i baed on a recurive algorithm. B (k) RAM Figure. The flowchart of auditory roughne analyi model. The auditory roughne level i dependent on the amplitude fluctuation and reult in level dependency on every frequencie. Thi correpond to the known definition of audio with different frequencie ( f and max f ) and amplitude ( min a a and a ) a follow. A Figure, According to STFT [7], Input ignal Xn ( ) into a local buffer, then take a length N FFT of Xn ( ) to obtain the STFT at window ize m : a a.. min, R a a. d( x) a a [] N j kn X ( ) X ( n) W ( n m) e [] nn where. fmax fmin.7 ( ) fmax f d x e e min ICA

3 -7 Augut, Sydney, Autralia Proceeding of th International Congre on Acoutic, ICA and..7 f 8.96 min The term a. a repreent the dependence of roughne on intenity of added inuoidal ignal, which i related to their amplitude. The term min a., a repreent the a a dependence of roughne on amplitude fluctuation degree of added inuoidal ignal, which i related to their amplitude difference. Finally, the term dx ( ) repreent the dependence of roughne on the amplitude fluctuation rate, which i related to frequency difference and depend on the frequency of lower inuoidal ignal. For every roughne reolution during the temporal domain, each roughne R ha to conider the unneceary amplitude fluctuation. Furthermore, the roughne of ignal correponding to pectra with more than two inuoidal ignal component can be calculated by umming the roughne of all inuoidal-pair in the pectrum. It i found that, depending on the relative phae of the repective amplitude fluctuation, the total roughne may be le than the um of the roughne value for individual pair. Therefore, the total roughne can be ummed over all auditory filter yielding a following, [] Hearing Balance and Hearing Satifaction Rank Hearing balance i the ratio of the total roughne pecific loudne H N ' Rt N ' a follow, R t and In the hearing of muic, not only the frequency difference, but alo the fluctuation in audio ignal, affect the hearing perception and ound quality. Eq. 8 indicate the fidelity and loudne in human enation. The Satifaction Rank H can be obtained over bark a follow. H a H [9] n n [8] T ' Rt T f fl ' t T b nf L R( n) where R t i total roughne according to every cale, t i the time range T T' of R, the frequency range f ha b been given in eq.. However, the ignal may vary in the temporal domain, the total roughne value from the frequency viewpoint need to conider non-ignal period. So, the total Roughne value mut be accurate over a pecific time period. Specific Loudne Loudne i a ound level perceptual meaure of the effect of energy content of ound on the individual ear which i related to ound preure level, but i not the ame. Loudne i the attribute of auditory enation in term of which ound may be ordered on a cale extending from oft to loud. Fundamental aumption of model of loudne indicate that it i not a product of pectral line and i not obtained from the pectral ditribution of the ound directly, but the total loudne i the um of pecific loudne from each critical band. Specific loudne N i given by N'( f).6 [ ( ) ] [6].ETQ. Ea ETQ. [7] where E TQ i the excitation at threhold in quiet, E i the excitation correponding to reference intenity (I O = W/m ), a i attenuation factor and E i the excitation level. Computer code in BASIC i provided by Zwicker []. Perceived ound quality of the ignal and it relation to the variou phyical propertie ha been the ubject of human litening comprehenion. where H n i hearing balance for particular bark cale. Eq. 8 etimate the atifaction rank/grade in the muic ignal. The coefficient a i deigned to upgrade the reult of Eq. 9. High and low frequency ound quality change are reflected in Eq. 8. The treble, middle and ba are often influenced by the audio ignal, hence we could ue Eq. 9 to obtain the average hearing balance and to oberve the Rank of ound quality of audio performance.paragraph immediately following their heading are to be jutified on both ide with no indent for firt line. Ue ingle line pacing throughout the entire document. There i a ingle line pace between paragraph. EXPERIMENT Five high fidelity headphone were elected judiciouly. Headphone were typical tudio headphone of variou tandard brand available in the market. Their cot variation i in the range of -8 US$ (Nov. 8). We have deignated them A to A. A i circumaural and emi-open headphone. A i circumaural and cloed-back headphone with autohut-off feature. A i circumaural and cloed-back headphone. A i circumaural and open tyle headphone and A i circumaural and cloed-back headphone. The obervable audio quality effect of thee headphone are ued for analyi and comparion of their audio quality reproduction. The experimental meaurement were carried out in an anechoic chamber by B&K electroacoutic equipment in accordance with the arrangement hown in Figure. We alo ued oftware SoundCheck 8.. The hardware connection were followed by calibration of device uing a B&K amplifier (type 76C) and a B&K tandard calibrator (type, khz ound at 9 db ±. db). B&K calibrator wa alo ued to calibrate B&K HATS imulator. A weep ine wave ignal from Hz up to khz (human audible frequency range) wa ued a an input to the audio ytem. Signal from amplifier wa recorded and proceed with SoundCheck 8.. Headphone were placed on B&K HATS for meaurement. Stimulu ignal (claical muic, the ong Somewhere in Time, conducted by Jeannot Szwarc, 98.) wa ued. Claical muic i often ditinguihed by it wide ue of intrument of varying tone and pitche to create a deep and rich ound. We have elected tet ignal in the range of : ~ :. Sound level did not change ignificantlty in thi range. ICA

4 Total Roughne (Rt) Total Roughne (Rt) Sone/ -7 Augut, Sydney, Autralia Proceeding of th International Congre on Acoutic, ICA Figure. The chematic diagram for record headphone. Specific loudne curve for each headphone along with reference ignal i hown in Figure. The overall pattern of thee curve i imilar, but it i till not exactly the ame over each bark. Thi wa found that the reference ignal in the high frequencie ( - 8 bark) i quite different from the other headphone ignal. It may be due to conideration of effect of hearing perception while recording. The reference ignal in digital form i upplied directly to analyer. However, the ignal from headphone may be affected by hearing enation due to HATS. Alo, it wa found that the pecific loudne in the low frequency region ( - bark) for headphone A wa lower than for other headphone, but in high-frequency region ( - 8 bark) pecific loudne wa higher than for any other headphone. Thi indicated the unique charatcterictic of A. The frequency repone for A i relatively flat, hence the muic reproduction would be more balanced. RESULTS AND DISCUSSION The Headphone Quality Comparion Figure illutrate the auditory roughne for each headphone along with the reference ignal uing ound quality aement model. Curve are imilar and nearly overlapping within low to mid frequency range (up to 7 bark), but above 7 bark, the change in auditory roughne i obervable. The enlargement of the above figure within the range of 6 - bark i hown in Figure. ignal auditory roughne i almot linear, however the performance of headphone varie from each other and even from the reference. It i totally pecific to performance, characteritic and deign. However, it i found that all auditory roughne curve converge to linear or near linear indicating a good ound quality performance. A A A A A Figure. Auditory Roughne for headphone along with reference ignal.. A A A A A A A A A A Figure. Specific loudne for different headphone. Figure 6 how the hearing balance for each headphone et with reference ignal. The atifaction rank of frequency performance with different audio compreion i hown in Table. The a 7.8 i defined in Eq. 9 a in the previou ection. Table how that the headphone A and A have higher atifaction rank, wherea A ha the lowet. Correpondingly, a hown in Figure 6, A and A have higher hearing balance among the five headphone within bark, wherea A ha the lowet. For the high frequency region ( bark), A and A repreent the wort and bet cae. Viual inpection of the hearing balance curve (more fluctuation) for A confirm that it i howing/reproducing irregular ound and appear to have ome rough or diordered ound. The performance of A and A eem to be much more regular. Similarly, A ha more creaky ound than A depite a higher atifaction rank for A. Figure 6 and Table how that the reference audio ha greater hearing perception when compared with the headphone.. Figure. Auditory Roughne for headphone along with reference ignal (- bark). ICA

5 Total Roughne (Rt) Hearing Balance -7 Augut, Sydney, Autralia Proceeding of th International Congre on Acoutic, ICA A A A A A Figure 6. Hearing balance of different headphone with reference ignal. Table. Satifaction rank with different headphone ound quality. Name Satifaction rank Refence A A A A A Audio Compreion Quality Comparion Figure 7 preent a general audio compreion coding flowchart which i a implification of MPEG Layer-III audio compreion model. The input ignal Xn ( ) and output ignal Yn ( ) are both mapped onto a pychophyical repreentation by mean of three operation, frequency warping, timefrequency mearing and compreion of the intenity cale. The making effect of pychoacoutic model predict the lole and loy of compreion rate which are baed on recontruction of audio ignal after FFT block [8, 9] and then encoded onto Yn ( ) which i affected by different making level. In our tet, the audio quality i influenced by controlled individual making threhold. a..7* z j c [] vtm where a i tonal making component, z(j) i the frequency index of the bark cale in which j i the index vtm label of tonal label. Treat the contant c a the variable for the purpoe of making the wrong tonal etimation of audio quality of ignal which i helpful to figure out good quality audio compreion. Comparion of the output of the audio compreion codec for making and un-coded making of reference ignal i good for undertanding the atifaction level of audio performance. A good reult for c i equal to. in the MPEG tandard []. Thi could be ued to explain the good atifaction rank in thi tet. Figure 8 how the auditory roughne curve with different audio compreion. Intead of a contant value for c, it effect on auditory roughne, pecific loudne and hearing balance were invetigated to figure out their importance. It wa found that by increaing variable c (higher compreion ratio) the audio quality became wore. Hence it can be aid that it i the wrong etimation of the final tonal making level. Alo, curve in Figure 8 preent the quality of audio fluctuation in the ignal. Thi indicate the timbre effect of the frequency range on hearing repone. It wa alo found that if the curve become nonlinear, the timbre woren. Thi i becaue the dionance repone of hearing perception are correlated with auditory roughne. The reference curve how the bet ound quality for original audio ignal and the balance of auditory roughne in every bark. For the value of c =, we oberve poor quality audio reproduction. c=. c=. c=. c=. X(n) X(n) FFT Pychoacoutic Model Enocoder Y(n) Y(n) Figure 8. Auditory roughne curve with different audio compreion. Audio Compreion Figure 7. Flow chart of audio compreion. The pychoacoutic model predict the final making level on the encoder. When making i exploited in audio bit rate reduction, the final making level make up a ignificant portion of the whole frequency-time pace. Enforcing variation in the making level enable detruction of or variety in the ubjective perceptual audio quality caue by the making threhold. The making index for tonal making component i given by Eq. [] with a contant c. The pecific loudne curve in Figure 9 how the variation in pecific loudne with different audio compreion. It i evident from the reult that the variation in c do not change the pecific loudne level. However, auditory roughne curve how ignificant change with thee variation. Specific loudne i not fully able to interpret the bet timbre for muic reproduction. A very mall change in the curve with c=. i oberved correponding to to bark (high frequency). Thi definitely repreent the ditortion of loudne due to audio compreion. ICA

6 Total Roughn (Rt) Hearing Balance Total Roughne (Rt) one/ -7 Augut, Sydney, Autralia Proceeding of th International Congre on Acoutic, ICA c=. c=. c=. c=. Figure 9. Specific Loudne with different audio compreion. The Optimum Auditory Roughne Curve A variety of muic ignal are directed to the analyzer in digital form for auditory roughne comparion. A long a the ound quality of ound ource i good, the reult hould be imilar and tend to be linear. From the analyi of four good CD-quality ong of different type in Figure (a), we found that reult are imilar and almot linear. Thi mean that auditory roughne alone can obtain an anwer a to the bet ound quality reproduction irrepective of the ound ignal type. For different kind of muic with good quality mix down and reproduction, by uing the regreion analyi, the bet reult for the roughne curve may be approximated by y.69 x.6 [] The hearing balance curve in Figure how how auditory roughne and pecific loudne collectively affect the hearing balance of ound quality. It i a pertinent obervation in the figure that with the increaing value in c, the curve depart from the reference curve at different bark. A we have already explained, the variation of pecific loudne with variation in c i very mall, hence the hearing balance i dependent on the ound fidelity only. In addition, change in audio compreion ratio ha the mot direct impact on the mid to high-frequency ignal range ( bark). where x range between correponding to bark. The reult of the above equation i hown in Figure (b). Claical Fuion Soul Rock-Metal.8.6. c=. c=. c=. c=. Figure (a). Auditory roughne curve - Four kind of muic.. Figure. Hearing balance of curve with different audio compreion. The atifaction rank of frequency performance with different audio compreion i hown in Table. The a 7.8 i defined in Eq. 9 of thi reult. A the ummation of hearing balance i.6, we did not make the atifaction rank higher than. If the atifaction rank decreae ignificantly, ound reproduction quality woren. If audio compreion goe beyond the limit (77 or below), ound quality will become very bad. Thi clearly ignifie that the ound become ditorted and dicordant. Table. Satifaction rank with different audio compreion. Name Satifaction rank Refence c=. 9. c= c= c=..9 Figure (b). Auditory roughne curve - Optimum curve. CONCLUSION Thi paper aeed ound quality baed on pychoacoutic. The reult clearly how that ubjective quantization method indicator can be ued to expre ubjective ound quality. The Auditory Roughne index can be ued to determine if, in term of frequency, i complete and whether fluctuation i exceive. In addition, Specific Loudne can ubjectively decribe if people reaction to loudne i balanced at all frequencie. It can ufficiently reflect opinion regarding 6 ICA

7 -7 Augut, Sydney, Autralia Proceeding of th International Congre on Acoutic, ICA ubjective repone to loudne. The ratio preent the integrated repone level of thee two auditory characteritic of Auditory Roughne to Specific Loudne may alo oberve the variation of ound quality. In our experiment, we compared ratio of audio compreion. Compreion rate aociated with good wound quality are known due to defined contant compreion ratio in ubjective apect. Thu, we can clearly oberve the level of completene of ound reproduction complete and the level of atifaction of auditory repone. Uing headphone a meaurement material to determine defect in ound reproduction fidelity and to implement improvement ha greatly contributed to the deign and quality management of headphone. Sound ignal evaluation method have greatly contributed to academic reearch and commercial development of peaker. The Sound Quality Model can be ued not only in evaluation of indutry but alo in evaluation of timbre muic performance. In thi tudy, although good and bad timbre repone are determined theoretically, quantitative difference cannot be acertained in muic preference among different ethnicitie. In future work, Comparion of ubjective teting and reult are uggeted with Jury Tet, uing a variety of muic and different ethnic group to obtain better Hear Balance hearing enation curve. ACKNOWLEDGEMENT The author would like to thank the National Science Council of Taiwan for financially upporting thi reearch under contract No. NSC 9--E---MY, 97--E-- -MY, and 96--E--MY. We alo thank Liten Inc. for upplying SoundCheck meaurement and Merry Electronic CO., LTD. for upplying headphone. REFERENCES P. H. W. Leong, Y. S. Moon, W. K. Sim, and D. W. P. Lam, Sound Quality Meaurement in Headphone, 6 th Audio Engineering Society (AES) Convention Munich, Germany, 87 (B6), pp. -, 8 - May 999. T. Hirahara, Phyical Characteritic of Headphone ued in Pychophyical Experiment, Acout. Sci. & Tech., Vol., No., pp 76-8,. K. K. N. Ravaja, Comparing Speaker Veru Headphone in Litening to New from a Computer Individual Difference and Pychophyiological Repone, Comput. Hum. Behav., Vol., Iue, pp. 7, Jan. 7 H. Y. Sung, J. Kim, S. Jang, A Method for Objective Sound Quality Evaluation of Headphone, AES nd International Conference, Hillerød, Denmark, pp. 8-87, September 7. T. Voinle and F. Briolle, Tranfer Function and Subjective Quality of Headphone: Part, Tranfer Function Meaurement, th International AES Conference, Audio Tet & Meaurement, Portland, Oregon, pp 8 -, 9- May H. Fatl, The Pychoacoutic of Sound-Quality Evaluation, Acutica, Vol. 8, No., Sept./Oct J. J. Chatterley, Sound Quality Analyi of Sewing Machine, Mater of Science Thei, Brigham Young Univerity, Department of Mechanical Engineering, Provo, UT, USA, Augut. 8 U. Zölzer, Digital Audio Signal Proceing, nd Edition, A John Wiley and Son Ltd, Publication, 8. 9 A. J. S. Ferreira, Perceptual Audio Coding and the Choice of an Analyi/Synthei Filter Bank and Pychoacoutic Model, th Audio Engineering Society (AES) Convention Amterdam, 67 (P9-), pp. - 6, 6-9 May998. J. G. Beerend, The Role of Informational Making and Perceptual Streaming in the Meaurement of Muic Codec Quality, AES the th Convention, 996 May -. Vailaki P. N., SRA: A Web-baed Reearch Tool for Spectral and Roughne Analyi of Sound Signal, Proceeding of, th Sound and Muic Computing Conference (SMC 7), Lefkada, Greece, pp. 9 -, - July 7. Vailaki P.N., SRA: An online tool for Spectral and Roughne Analyi of ound ignal, J. Acout. Soc. Am., (/): 677. E. Zwicker and E. Terhardt, Analytical Expreion for Critical-Band Rate and Critical Bandwidth a a Function of Frequency, J. Audio Eng. Soc., Vol. 68, No., pp. -, Nov. 98. E. Zwicker, H. Fatl, U. Widmann, K. Kurakata, S. Kuwano and S. Namba, Program for Calculating Loudne According to DIN6 (ISO B) J. of Acout. Soc. of Jpan. (E), Vol., No., pp. 9 -, 99. J. E. Appell, Loudne Model for rehabilitative Audiology, Doctor of Science diertation, Univerity of Oldenburg, Germany, February. 6 R. M. Aart, Calculation of Loudne of Loudpeaker during Litening Tet, J. Audio Eng. Soc., Vol. 9, No. /, pp. 7-8, Jan./Feb J. O. Smith and X. Serra, "PARSHL: An Analyi/Synthei Program for Non-Harmonic Sound Baed on a Sinuoidal Repreentation," Proceeding of the 987 International Computer Muic Conference, International Computer Muic Aociation, San Francico, pp C. M. M. Tio, R. K. C. Tan and C. K. Yeo, Pychoacoutic Parametric Audio Coding Technique Uing a High-Order QM Sub-Band Filter Bank, 7 th Audio Engineering Society (AES) Convention, New York, 6 (N- ), pp. - 8, - 7, September R. Sušnik, J. Sodnik, S Tomažič, An Elevation Coding Method for Auditory Diplay, Appl Acout., Vol. 69, Iue, pp. -, March 8. ISO/IEC 7-, Information technology -- Coding of moving picture and aociated audio for digital torage media at up to about, Mbit/ -- Part : Audio, International Standard for Buine, Government and Society, 99. ICA 7