In Order to establish a comprehensive method of evaluating reproduced sound quality, models of the process of sound quality evaluation were proposed. It was assumed that the total process of sound quality evaluation could be divided into two processes, an overall (emotional) process and an elemental sensory process, that formed a sensory factor of the former process. The elemental sensory process is described in terms of a multidimensional scale such as D_i=φ_i(t_j, s_i)(i=1, 2. . . , n), where t_j: a physical parameter showing the characteristics of a transmission system, s_i: components of S contributing to D_i. The elemental sensation D_i is assumed to have such relation with overall (emotional) response, R, (in reality, preference scale), as R=Σw_iD_i, where w_i=W(L, S, A), L: listener groups classified in terms of the similarity of preference, S: musical signal groups classfied by their effect on w_i, and A: age (time). The methodology to realize those models, especially that of the elemental sensory process, was discussed. According to this discussion, multidimensional scaling of the sensation of reproduced sound quality was made, varying the klirrfactor and the high-pass and low-pass filtering frequencies of transmission characteristics, as shown in Table 4. Fig. 15-17 show the three-dimensional scale thus obtained. In order find more definite relation between D_i and t_j, two more experiments were made. Table 1 shows the stimulus condition of an experiment on D_1. The inputoutput characteristics of a nonlinear distortion circuit are shown in Fig. 4. As shown in Fig. 5 and 6, a two-dimensional scale was calculated. But it was found that one-dimensional variation was only significant. Fig. 9 shows the relation between D_1, projections of the stimulus on the rotated unidimensional axis (dotted line) in Fig. 5 and 6 and the klirrfactor of the transmission system. Another experiment was concerned with D_2 and D_3. Stimulus condition is shown in Fig. 10 graphically. The two-dimensional centroid scales and rotated scale thus obtained are shown in Fig. 11 and 12. Fig. 13 and 14 show the graphic presentation of functions φ_2 and φ_3. These experiments only serve as the illustrations of our models. Several other experiments have been done, and from these experiments, we have found that five dimensions are enough for describing major sensory changes that will occur with usual variations in the physical characteristics of the transmission system.

View full abstract

A model of the process of evaluating reproduced sound quality was examined experimentally. As shown in Fig. 1, it was assumed that there was such relationship between the preference scale of sound quality R and multidimensional sensory scale D_i(i=1, 2. . . , n), as R=w_iD_i, where w_i=W(L, S, A), L: listener groups classfied in terms of the similarity of preference, S: musical signals classfied by their effect on w_i, A: age(time) and D_i=φ_i(t_j, s_i), t_j: a physical parameter of a transmission system, s_i: components of S, contributing to D_i. To examine this model, three experiments were made. The first and the second experiments were made to examine a linear equation relating to above-mentioned R and D_i. Stimulus conditions are shown in Table 1, and the block diagram of experimental circuits is shown in Fig. 2. Two of classic music(A: synphony, B: string quartet) were used as the source S, and presented to ten listeners under the above stimulus conditions. By means of the method of paired comparisons, the preference scale R and two sensory scales D_1(sensation of noises, mainly related to nonlinear distortion) and D_3(sensation ob treble, mainly related to low-pass filtering frequency) were obtained. From three scales the least square solutions of w_1 and w_3 were calculated. These values are shown in Fig. 3 and 4. Finally, from the scale values of D_1 and D_3 and estimated weights w_1 and w_3, estimated value R^^~ of preference scale was calculated. As shown in Figs. 3 and 4, the coincidence of observed value R and estimated value R^^~ was fairly good. This seems to prove the adequacy of the model. Fig. 5 shows the coincidence of R and R^^~ with four dimenstional sensory scales (D_1-D_4) taken into account. This third experiment was made to examine the stability of w_i. If w_i is only affected by L, S and A, and not by t_j, and once w_i for definite L and S is obtained, as far as L and L and S are fixed, R^^~ for various t_j will be able to be estimated from the curves of φ_i. These estimated values of R^^~ should coincide with observed R for such t_j. Fig. 6 shows the curves, in which w_i was calculated, and Fig. 7 shows tho estimated scale values of D^^~_<2i> for the second set of t_j, which is different from that used in determing w_i (the first set of t_j). From such predetermined w_i and estimated D^^~_<2i>, estimated value R^^~_2 of preference scale was calculated. As shown in Fig. 8, estimated value R^^~_2 and obserbed value R_2 for the second set of t_j coincided with each other very well again. If this model is proved to be adequate, and when a listener group and a source classification are given, it will became possible to design a transmission system, the sound quality of which is liked best by the listener group.

View full abstract

Temporary threshold shift (TTS) at 1, 2, 3 and 4 kc following exposure to four types of octave-bands (250-500cps, 500-1000cps, 1000-2000cps, 2000-4000cps) obtained by passing white noise through filters was measured with five subjects with normal hearing acuity. Table 1 gives the sound pressure levels of octave-band noises used in this experiment. Durations of exposure ranged from 5 to 155 minutes, and measurements of TTS were made within 3 minutes after cessation of exposure. (Fig. 3) Variations of pre-exposure thresholds of the five subjects are shown in Table. 2. All experiments were carried out in an anechoic soundproof room. (Fig. 1) Fig. 4 shows the TTS produced by four different octave-band noises at a level of 95dB. Among the four test frequencies (1, 2, 3 and 4kc), maximum effect was found at 3kc for 1000-2000cps octave-band, and at 4kc for 2000-4000cps octave-band. As regards the bands 250-500cps and 500-1000cps, TTS at 1, 2, 3 and 4kc decreased in that order. The relations of TTS to exposure time and sound pressure level are shown in Fig's 5 and 6. As Ward et al. had already pointed out, TTS increased linearly with the sound pressure level of exposure noise and the logarithm of exposure time. These relastion are expressed as TTS=a(S+b)log_<10>T+cS+d where S denotes the sound pressure level of each octave-band noise, T is exposure time in minutes, and a, b, c and d are constants that depend on octave-band and test frequency. These constants were calculated by the least square method. The empirical formulas obtained are shown in the text. (from (1) to (8)) Fig. 7 indicated the correspondence of the original data to a regression equation for TTS at 4kc following exposure to 2000-4000cps noise. Comparing the equations of Ward et al. 's with those of authors', TTS at 4kc due to 2000-4000cps noise was larger than that due to 2400-4800cps noise, (Ward et al. ) and TTS at 3kc due to 1000-2000cps noise were smaller than that due to 1200-2400cps noise (Ward et al. ). (Fig. 8) Fig. 9 shows the sound pressure level of octave-band noise which give, for 3 minutes exposure, TTS of 5dB at frequency half an octave above the center frequency of the octave band noise. Broken line represents the results obtained by Plomp et al. , and solidline is from the present equations. They are in good agreement.

View full abstract

In order to investigate whether the critical band concept can be applied to the problem of temporary threshold shift (TTS) , three experiments (I, II, III) were carried out using five subjects with normal hearing acuity. In experiment I, thirteen high pass and thirteen low pass noises obtained by filtering white noise were used. The cut-off frequency of these noises are shown in Fig. 1. They were at intervals of 1/6 octave. The over-all SPL of white noise was 95dB. Durations of exposure were 5, 15, 35 and 55 minutes, and post-exposure threshold measurements at 3, 4, 6 and 8kc were made whithin 3 minutes after cessation of exposure. Fig. 2 shows the results of experiment. TTS due to low pass and high pass noises increased to a certin value as the bandwidth became larger, but when it reached to this limiting value, it remained constant regardless of the bandwidth of exposure noise. It may be concluded from this fact that only those components of the noise which are included in limited frequency regions are effective and that the components beyond this regions are ineffective in TTS. This is in agreement with the basic notion of the critical band. In experiment III, twelve exposure noises having linear spectrum were used (Table 1). The spectra of these noises are given in Fig. 3. TTS at nine frequencies from 0. 5kc to 8kc were measured within about 6 minutes after cessation of 20 minutes' exposure. Fig. 4 shows the TTS due to exposure to these noises at a level of 100dB. As a whole, 0dB/oct noise was most effective and -6dB/oct noise least effective. But TTS at frequencies below 2kc were not noticeable in all cases. In experiment III, four 1/6 octave-band noises (2240-2500cps, 2800-3150cps, 4500-5000cps, 5600-6300cps) whose spectrum level are equal to that of 0dB/oct noise at 100dB were used. Test frequencies and exposure time were the same as in experiment II. Fig. 5 indicates the results of this experiment. The maximum effects were found at 3, 4, 6 and 8kc respectively for the exposure noise 2240-2500cps, 2800-3150cps, 4500-5000cps, and 5600-6300cps. Using the data obtained from experiment II and III, the center frequency and width of the critical band were estimated by the following method. 1) Estimation of the center frequency of the critical band. The basic assumption is that TTS at frequency F is expressed as TTS_F=aX+b. . . . . . . . . . (1) a, b: Constants that depend on exposure time, test frequency, and the time when TTS is measured. X: Critical band level and is expressed as X=S(F_c)+10log_<10>&lrtri;f. . . . . . . . . . (2) S(f_c): Spectrum level at the center frequency of the critical band f_c: Center frequency of the critical band &lrtri;f: Critical bandwidth When the spectrum of noise is a linear function of log_2f, S(fc)=αlog_2f_c+β. . . . . . . . . . (3) α: Spectrum slope (dB/oct) β: intercept (dB) From Equations (1), (2) and (3), TTS_F=a(αlog_2f_c+β-L). . . . . . . . . . (4) where L≡-(10log_<10>&lrtri;f+b/a) Equation (4) means that TTS becomes a linear function of the spectrum level at the center frequency of the critical band. Using the data of experiment II, the value of a, f_c, and L were calculated for 3, 4, 6 and 8kc by the following least squre method. &lrtri;=Σ{y_i-a(α_ilog_2f_c+β_i-L)}^2 ∂&lrtri;/(∂a)=0, ∂&lrtri;/(∂f_c)=0, ∂&lrtri;/(∂L)=0 where y_i is TTS produced by noise whose spectrum is α_ilog_2f+β_i. The results are shown in Fig. 6. From these figures, it is noticed that TTS is expressed as a linear function of spectrum level at the center frequency of the critical band. Center frequencies are about one-third to two-third octave below test frequencies. 2) Estimation of the critical bandwidth. Let the TTS at certin frequency produced by wide-band noise (I) in Fig. 7 be Y, and the TTS by narrow-band noise (II) whose cut-off frequencies are included in the critical band be y, then

(View PDF for the rest of the abstract.)

View full abstract