Effect of room acoustic conditions on masking efficiency

Effect of room acoustic conditions on masking efficiency Hyojin Lee a, Graduate school, The University of Tokyo Komaba 4-6-1, Meguro-ku, Tokyo, 153-855, JAPAN Kanako Ueno b, Meiji University, JAPAN Higasimita 1-1-1, Tama-ku, Kawasaki-si, Kanagawa, 214-8571, JAPAN Shinichi Sakamoto, Institute of Industrial Science, The University of Tokyo Komaba 4-6-1, Meguro-ku, Tokyo, 153-855, JAPAN Mai Fujiwara c, Yasushi Shimizu, Masato Hata Center for Advanced Sound Technologies, YAMAHA Corporation 3 Matsunokijima Iwata-si, Shizuoka, 438-192, JAPAN ABSTRACT On the evaluation of a sound masking system in an actual room environment, not only the signal property of masking sounds (maskers) but the room acoustic condition gives a significant influence on masking efficiency. In this study, the effect of the room acoustic conditions on the masking efficiency is experimentally investigated by simulating three dimensional impulse responses of transmission property from sound sources (talker and masker) to a listener. The experiment is designed to examine the influence of the following three components of the room acoustical characteristics; frequency characteristics, transient characteristics and spatial characteristics. The degree of the masking efficiency by a noise masker and a mix masker (noise and speech-like sound) is measured by a word intelligibility test for a talker s voice transmitted through walls. The tests are conducted in an anechoic room with a three dimensional sound simulation system to reproduce the spatial characteristics of the actual rooms. From the experimental results, the effect of each room acoustical characteristics on the masking efficiency is measured. In addition, the result indicates that the masking efficiency of the mix masker is superior to the noise masker, which is supposed to be caused by the effect of information masking. a Email address: leehj@iis.u-tokyo.ac.jp b Email address: uenok@isc.meiji.ac.jp c Email address: mai_fujiwara@gmx.yamaha.com

1. INTRODUCTION A sound masking system is a device that is used to prevent a third party from hearing oral information that is desired to be kept private. This system is very useful in the spaces where sound insulation is insufficient. Recently, speech privacy [1] has become a topic of high interest in facilities that is required to secure personal information. There have been many studies on the evaluation methods of the sound masking system. Indices calculated from signal-to-noise ratio at each frequency band are the most prominent example of the evaluation technique. Privacy index [2], Speech intelligibility index [3] and SNRuni32 [4] for example, are the main speech privacy indices considering frequency characteristics. In our past studies [5-6], for the evaluation of the masking efficiency in actual use, we had reproduced such transmission properties caused by room acoustic conditions as the deterioration and the direction of the test sounds in actual acoustic environment using a 3-dimentional sound simulation system [7-8]. From the result, we had observed that the transmission properties had a significant effect on the masking efficiency. However, it was not clear which components of the transmission properties had changed the masking efficiency. In this study, the transmission properties in the frequency domain (change of the frequency characteristic of sound caused by the insulation performance of walls), in the transient domain (change of the transient characteristics caused by reverberation of the space) and in the spatial property (locations of the target, masker and background noise) were focused on and examined each characteristics contribution on masking efficiency. 2. EXPERIMENTAL METHOD A. Sound Field Model Sound field model with the sound masking system was composed of the target (masked speech), the masker (masking sound) and the background noise. Figure 1 shows 4 test cases set in the experiment. Frequency characteristics SPL [db] SPL [db] SPL [db] SPL [db] Freq. [Hz] Freq. [Hz] Freq. [Hz] Freq. [Hz] Transient characteristics Time [sec] Time [sec] Time [sec] Time [sec] Spatial characteristics Listener Listener Listener Listener a. Case-none b. Case-F c. Case-F+T d. Case-F+T+S Figure 1: Sound field model of components on experimental conditions

Case-none (see Figure 1a): This case was set to examine the situation in which the acoustical characteristics was not considered. The target and masker were provided by dry sources. The background noise which was measured in the real room was used. All of the test sounds were reproduced with a monaural loudspeaker. Case-F (see Figure 1b): This case was set to examine the influence of the frequency characteristics of the target and masker being changed by the room acoustic conditions. In addition, the background noise which was measured in the real room was provided. The frequency characteristics of the target, masker and background noise used in the experiment were adjusted to the ones measured in the real room (see section 2-C). All of the test sounds were reproduced with the monaural loudspeaker. Case-F+T (see Figure 1c): This case examined the influence of the transient characteristics (such as reverberation) of the room. The target and masker were convolved with a monaural impulse response. In addition, the background noise which was measured in the real room was provided. The frequency characteristics of the three test sounds were set to be the same as those in Case-F. All of the test sounds were reproduced with the monaural loudspeaker. Case-F+T+S (see Figure 2d): This case examined the influence of the spatial characteristics of the test sounds. The locations of the target, masker and background noise in the real acoustic condition were simulated in the experimental space. The transient and frequency characteristics of all the test sounds were adjusted to the same as Case-F+T. All of the test sounds were reproduced by a 3-dimensional sound field simulation system (see section 2-B). B. Sound Reproduction System Background Sound CD Player (2ch) Hard Disk Recorder (6 ch Recorded sound) Level Controller Digital Mixer Yamaha DME64 Real-time Convolver (IRM, 6 ch.) Level Controller Mixer Real-time Convolver (IRT, 6 ch.) Level Controller Equalizer for loudspeakers DA Converter (6 ch) 2, mm Power Amp. to 6 ch. Loudspeakers set in anechoic room seating direction 2 seating direction 1 In anechoic room Figure 2: 3-dimensional sound field simulation system in anechoic room In order to create the condition of case-f+t+s, the 3-dimensional sound field simulation system was employed to simulate actual sound environments in an anechoic room with 3-D information. As reproducing signals in this system, sounds recorded in an actual field through a microphone system comprising six unidirectional microphones (Sanken, CU-6ch) developed for this reproduction system were used. The 6-ch. signals recorded in the actual field were reproduced

through six loudspeakers (TANNOY, T12) arranged on a spherical surface with a radius of 2 m in the anechoic room. Using this system, a subject at the center position of the system obtained a natural 3-D auditory sensation. Figure 2 shows the diagram of the sound reproduction system for the target, masker and background noise. In Case-none, Case-F and Case-F-T, test sounds were reproduced through one of the six loudspeakers set in front of the subject who was seated toward to the direction 1. In Case-F+T+S, the 3-dimensional sound field simulation system was used and the subject seated toward to the direction 2. To provide the transmission properties of the target and masker in case F+T and case F+T+S, the impulse responses, IR T (for the target) and IR M (for the masker), in the acoustical room condition were convolved using a real-time convolution system. In Case-F+T+S, 6-ch. impulse responses were used. The background noise was reproduced by a hard disk recorder. The levels of presented sounds were adjusted by the A-weighted SPL at the center position of the simulated sound field for all experimental conditions. In the listening tests, the subject was seated at the center position of the experimental sound field and listened to the test sounds. C. Simulated room conditions Two kinds of test fields with different room acoustic conditions were set (see Figure 3). Room 1 was a condition that two rooms were connected with partition walls and Room 2 was a condition that a single room was surrounded by high partitions in a large space. Both rooms acoustic conditions were insufficient to secure speech privacy. In order to reproduce the sound field for the subjective experiments, the acoustic measurements of the impulse responses and the background noise were conducted. Figure 3 shows the dimension of the spaces and the settings of the sound sources and the receivers (listeners). Four loudspeakers were set at the corners to reproduce the masker. sliding door high partition (h=2.2m) 3.9 m 3.4 m Receiver (listener) Receiver (listener) 2.15 m 2.15 m 2.6 m (a) Room 1 (b) Room 2 Figure 3: Dimension of spaces and measurement positions in the each room Figure 4 shows the frequency characteristics of the target in the two rooms under four experimental conditions. Figure 5.a and Figure 5.b were the frequency characteristics of masking sound that consists of the MIX (see section 2-D) masker (at 45 db) and background noise (at db) under four experimental conditions in the two room conditions. Figure 5.c and Figure 5.d were the ones for the NOISE (see section 2-D) masker (at 45 db) and background noise (at db).

SPL [db] 5 3 1 125 25 5 1k 2k 4k 8k 125 25 5 1k 2k 4k 8k (a) target at Room 1 (b) target at Room 2 Case-none Case-F Case-F+T Case-F+T+S Figure 4: Frequency characteristic of the target of four cases in the each room SPL [db] SPL [db] 5 3 1 5 3 1 125 25 5 1k 2k 4k 8k (a) MIX (45 db) + BGN ( db) at Room 1 125 25 5 1k 2k 4k 8k (c) NOISE (45 db) + BGN ( db) at Room 1 125 25 5 1k 2k 4k 8k (b) MIX (45 db) + BGN ( db) at Room 2 125 25 5 1k 2k 4k 8k (d) NOISE (45 db) + BGN ( db) at Room 2 Case-none Case-F Case-F+T Case-F+T+S Figure 5: Frequency characteristic of the maskers (at 45 db) and BGN (at db) of four cases in the each room D. Test Conditions Table 1 shows the level settings of the test sounds. The target and background noise were db in A-weighted SPL. With the regard to the target, five mora words with the same wordfamiliarity were chosen from the NTT Database Series: Lexical Properties of Japanese [9]. The words were recorded by a single female. As the masker, two maskers were used in the test, one was a steady-state noise masker: NOISE, the other one was a mixed sound of speech-like sound and steady-state noise: MIX. SPL of masker was changed in 5 db steps. MIX was changed from 35 to 5 and NOISE was changed from to 55. As the background noise, the recorded air conditioning noise was used The experimental conditions were the combination of three cases (Case-F, Case-F+T, Case- F+T+S) with two masker types, four masker levels and two room types. The total number of conditions was 56. In addition, the conditions (Case-none) for using original dry sources of target and masker with two masker types and four masker levels, 8 conditions in total, were examined.

Table 1: Level settings of the test sounds Test sounds Level* [db] MIX 35,, 45, 5 NOISE, 45, 5, 55 Background noise * A-weighted sound pressure level E. Procedures The masking efficiency was measured by means of the word intelligibility test. As the target, eight words were presented one by one with a four-second blank, and subjects wrote down the presented words in the blank. The subjects were encouraged to guess the words. Nine subjects in their -3 s with normal hearing ability participated in the experiment. 3. EXPERIMENTAL RESULT The mean of the percentages of correct responses of the word intelligibility test for each subject was calculated for each condition and the analysis of variance was conducted with experimental case (Case-none was not included), room type, masker type and masker level as factors. From the ANOVA, it was found that the main effects of the case type, room type, masker type and masker level were highly significant and some interactions were detected (see Table 2). The coefficient of determination in the ANOVA model was 93 %. Furthermore, in order to investigate the difference between conditions of the case type, a multi-pair comparison test (Tukey s HSD test) was conducted. From the results, all of the six pairs composed by the combination of three cases were significant (p<.5). Table 2: Result of the analysis of variance Source of Variation The degree of Sum of freedom squares F-value p(prob>f) Case type 2.4 33.516 <.1 Room type 1.481 79.954 <.1 type 1.436 72.48 <.1 level 1 1.641 272.641 <.1 Case type * Room type 2.61 5.41.13 Case type * level 2.11 8.363.1 Case type * type 2.68 5.67.8 Room type * type 1.49 8.189.7 Room type * level 1.26 4.367.45 Room type * level * Case type 2.39 3.195.54 For each masker type and level, the word intelligibility test score (vertical-axis) was plotted against the case type (horizontal-axis) in Figure 6. Error bars show 95 percent confidence intervals. High score of the word intelligibility test meant that the masking efficiency was poor. The results of the experiment were listed as follows. The intelligibility score of each case type was varied. Regardless of the cases, when the masker level was increased, the masking efficiency became better.

1 8 level= 35 SNRuni32 * =-2. SII * =.38 level= SNRuni32=-3.37 SII=.35 level= 45 SNRuni32=-5.63 SII=.28 level= 5 SNRuni32=-9.1 SII=.18 Word intelligibility test score [%] 6 1 8 6 level= SNRuni32=-3.21 SII=.37 (a) MIX level= 45 SNRuni32=-5.19 SII=.31 level= 5 SNRuni32=-7.91 SII=.23 level= 55 SNRuni32=-11.68 SII=.15 [none] [F] [F+T] [F+T+S] [none] [F] [F+T] [F+T+S] [none] [F] [F+T] [F+T+S] [none] [F] [F+T] [F+T+S] Case type (b) NOISE MIX NOISE *SNRuni32 and SII were calculated for Case-F Figure 6: Word intelligibility score of the four test conditions for Room 1 Even in the cases with the same frequency characteristics of the test sounds, the masking efficiency considerably varied by the influences of the transient characteristics and the spatial characteristics. This result indicated that considering only the frequency characteristics was insufficient for assessing the masking efficiency in realistic room sound conditions. Comparison between Case-F+T and Case-F+T+S revealed that the spatial characteristics made the masking efficiency low. This result indicated that the segregation of the test sound s location led to the decrease of the masking efficiency. This phenomenon was generally called spatial unmasking [1]. Comparing the results between the MIX and NOISE maskers, the level of the MIX masker tended to be 5 db lower than that of the NOISE masker at the same word intelligibility score. This result indicated that the MIX masker was more efficient than the NOISE masker. The fluctuation contained in the MIX masker seemed to have positive effect on the masking efficiency. Case-none was not included in the analysis of variance that was conducted to confirm the effect of factors on the masking efficiency. However, the Figure 6 showed that the effect of the frequency characteristics was confirmed by comparing the results between Case-F and Case-F+T

Figure 7(a) and 7(b) show the comparison between Case-F and Case-F+T for Room 1 and Room 2, respectively. The intelligibility scores of the MIX masker were plotted versus the masker level of two rooms. The decrease of the intelligibility score by adding the transient characteristics is more significant at the Room 1. In order to examine the difference between two rooms from the acoustic viewpoint, reverberation time and room acoustical indices obtained from the impulse responses at 5-1k Hz were calculated (see Table 3). These results indicated that clarity of the target sound significantly affected on the masking efficiency. Word intelligibility test score [%] 1 8 6 35 4 5 5 45 5 55 level [db] level [db] (a) Room 1 (b) Room 2 Case-F Case-F+T Figure 7: Word intelligibility test score of the MIX masker in Case-F and Case-F+T in the each room Room 1 Room 2 Room type T 6 [s] D 5 [%] C 5 [db] T s [ms] 1 3 Time [ms] Figure 8: Waveforms of the impulse responses of Case-F+T in the each room Table 3: Acoustic indices of Case -F+T Room 1.62 57.3 1.1 56.3 Room 2.44 94.4 14.24 4. CONCLUSION This study investigated the effect of the room acoustical characteristics, the frequency characteristics, transient characteristics and spatial characteristics, on the masking efficiency using the word intelligibility test. From the experimental results, it was clarified that each characteristics had a significant influence on the masking efficiency. The influence of each characteristics was summarized as follows. The transient characteristics caused by the room acoustic condition made the sound masking efficiency high. The amount of the increase of the sound masking efficiency was dependent on the room acoustic condition. The spatial segregation of the test sounds by the spatial characteristics induced the decrease of the masking efficiency. The masking efficiency was also affected by the masker level and the masker type. The frequency characteristics of the test sounds had an effect on the masking efficiency. However, even if the test sound had the same frequency characteristic, the masking efficiency 14.

was changed by other acoustical characteristics such as the transient characteristics and the spatial characteristics. The results obtained in this study indicated that the three kinds of room acoustical characteristics should be considered for the evaluation of the masking efficiency. In order to generalize each effect, further investigation is required. REFERENCES 1 W. J. Cavanaugh, W. R. Farrel, p. W. Hirtle, and B. G. Watters, Speech Privacy in Buildings, Journal of the Acoustical Society of America, 23, 475-492 (1962). 2 American Society for Testing and Materials, ASTM International, E113-2e1, Standard test method for objective measurement of speech privacy in open offices using articulation index, ASTM international, west Conshohocken, Pa. (2) 3 American National Standard Acoustical Terminology, American National Standards Institute ANSI S3.5 (Acoustical Society of America, New York, 1997). 4 Bradley, J. S. and Gover, B. N., A new procedure for assessing the speech security of meeting rooms, proceedings institute of Acoustics, U. K., v., 1-6 (8) 5 A. Ito, A. Miki, Y. Shimizu, K. Ueno, HJ. Lee, and S. Sakamoto, Oral information masking considering room environmental condition, Part 1: Synthesis of s and examination on their masking efficiency,, proceedings of INTER-NOISE 7 (7). 6 K. Ueno, HJ. Lee, S. Sakamoto, A. Ito, A. Miki, and Y. Shimizu, Oral information masking considering room environmental condition, Part 2: Subjective assessment for Masking efficiency and Annoyance,, proceedings of INTER-NOISE 7 (7). 7 S. Yokoyama, K. Ueno, S. Sakamoto, and H. Tachibana, 6-channel recording/reproduction system for 3- dimensional auralization of sound fields, Acoustical Science and Technology, 23(2), 97-13 (2). 8 S. Yokoyama, H. Yano, and H. Tachibana, 6-channel recoding/reproduction system for 3-dimensional auralization and it s applications to psycho-acoustical experiments, proceedings of INTER-NOISE 7 (6). 9 S. Amanao and T. kondo, Nihongo no goitokusei [Lexical properties of Japanaese] (Vols.1-6), NTT database series, Tokyo, (1999). 1 Richard L. Freyman and Karen S. Hlfer. The role of perceived spatial separation in the unmasking of speech, Journal of the Acoustical Society of America, 16(6), 3578-3588 (1999)