A METHOD FOR BINAURAL SOUND REPRODUCTION WITH WIDER LISTENING AREA USING TWO LOUDSPEAKERS

Transcription

1 23 rd International ongress on Sound & Vibration Athens, Greece 0-4 July 206 ISV23 A METHOD FO BINAUA SOUND EPODUTION WITH WIDE ISTENING AEA USING TWO OUDSPEAKES Keiichiro Someda, Akihiko Enamito, Osamu Nishimura and Takahiro Hiruma Toshiba orporation, orporate esearch & Development enter, Kawasaki, Japan keiichiro.someda@toshiba.co.p rosstalk cancellation is a technique for binaural sound reproduction using two loudspeakers. rosstalk cancellation filters are formulated for a binaural sound to be reproduced at the positions of the ears of a listener. The drawback of crosstalk cancellation is the narrowness of the listening area, that is, the quality of the reproduction is sensitive to the movement of the head. A new method that achieves wider listening area is proposed. In this method, filters are formulated for not the binaural sound itself but its interaural complex sound pressure ratio (IA) to be approximately reproduced at plural listening positions. It is assumed that the listener can perceive a virtual sound source in the nearly identical direction by reproducing the IA. ompared with trying to reproduce the binaural sound itself, the error amount of the reproduction is small in the case of trying to reproduce the IA at the listening positions. The sound image the listener perceives is almost unchanged even if about 0 centimeters head movement occurs, so the head positioning is easier than that of crosstalk cancellation. The reproduction demonstrations have held using loudspeakers below a TV display, USB loudspeakers connected to a personal computer, and built-in loudspeakers of a personal computer, a handheld game console, and a tablet computer, for many listeners to enoy the virtual sound effects of the binaural sounds.. Introduction The fundamental idea of 3D sound reproduction is to deal with the acoustical signals at the listener s ears, and thereby reproduce a particular auditory scene. These systems can be used in various fields, such as video conferencing, home entertainment, 3D games, virtual reality, air traffic control, and pilot warning systems []. There are two ways to deliver binaural sound signals to the ears. One way is through headphones. The other is by using loudspeakers. When loudspeakers are used to deliver binaural sound signals, the "crosstalk" signal that arrives at each ear from the opposite-side loudspeaker should be eliminated. rosstalk cancellation system (SS) is an efficient way to eliminate these "crosstalk" signals. In the traditional Atal-Schroeder crosstalk canceller [2], the two loudspeakers are placed. It is well-known that such systems are only effective when the listener is in a known position, therefore, the so-called "sweet spot" is too small. That is to say, the filters are designed to give perfect reproduction with a predetermined head position of the listener; then the reproduction is distorted when the head moves. In [3], tracking the listener s head position is proposed to constantly update the filters to maintain good reproduction. However, it is not always feasible to ask the listener to wear the tracker everywhere. opez and Gonzalez [4] deals with the robustness of cross-talk cancellation against slight movements of the listener s head regarding the loudspeakers angle of listening. But using its knowledge loudspeakers location cannot be determined freely.

2 The 23 rd International ongress of Sound and Vibration In this paper, to achieve binaural sound reproduction with wider listening area using two loudspeakers, we focus interaural complex sound pressure ratio (IA) of binaural sound. The IA of binaural sound is defined by, IA s = s s () where s and s are the complex signals of the left and right channel of the binaural sound, respectively. Similarly, the IA of the sound listener receiving is defined by, IA P = P P (2) where P and P are the complex sound pressures at the left and right ear of the listener, respectively. It is assumed that the listener can perceive a virtual sound source in the nearly identical direction when the IA of binaural sound is reproduced at the ears of the listener, that is, IA P = IA s is satisfied, since IA contains whole information of both interaural time difference (ITD) and interaural level difference (ID) of binaural sound, which are cues for listener to perceive the direction of sound source. The limitation of this assumption is that vertical localization cannot be realized in theory because it mainly depends on monaural localization cue. In the proposed method, filters are formulated for the IA of binaural sound to be approximately reproduced at pairs of both ear positions of plural listening positions, while crosstalk cancellation tries to reproduce sounds of both left and right channels of binaural sound separately at positions of left and right ears of listener, respectively. 2. Proposed method A control law for approximating the IAs of the sound pressures at listening positions of arbitrary number N to the IA of the sound arriving at both ears of listener from a virtual sound source, using loudspeakers of arbitrary number M, is derived hereinafter. Fig. shows the reproduction configuration under discussing. Here, i is the index corresponding to listening positions (i = to N), and S W W 2 W i W M q q 2 q =~M q M i i S d d i i i N d i Virtual 仮想音源 Sound Source d i P P i=~n Enlargement 聴取位置の拡大 of istening Area Figure : Sound field reproduction by trans-aural system with Multi-channel loudspeakers. is the index corresponding to loudspeakers ( = to M). Positive integer M represents the number 2 ISV23, Athens (Greece), 0-4 July 206

3 The 23 rd International ongress of Sound and Vibration of loudspeakers and M 2. Positive integer N represents the number of listening positions and N. P i and P i represents complex sound pressures at the left and right ear of the i-th listening position, respectively, i, i represents a head related transfer function (HTF) from the -th loudspeaker to the left and right ear of the i-th listening position, respectively, q represents an output sound signal of the -th loudspeaker, W represents the control filter coefficient of the -th control filter, and s represents a monaural sound source signal. d i and d i represents a HTF from a given virtual sound source to the left and right ear of the i-th listening position, respectively. If the absolute position of the virtual sound source is the same at each listening position as shown in Fig., d i and d i change with listening position. But here we hope its position relative to listener stay unchanged wherever in listening area, therefore we set the listener perceiving the same virtual sound source at every listening positions, that is, d = d 2 =... = d N = d, (3) d = d 2 =... = d N = d. (4) The aim of this reproduction is that IA P = IA s is satisfied at each listening position, that is, P i P i = d i d i To achieve this, Q of Eq.(7) should be minimized: = d d ( i N). (5) P i = d P i d P i ( i N), (6) N Q = P i P i. (7) i= The condition for Q to be minimized is: Q W = 0 ( M), (8) therefore, by doing this partial differentiation with respect to each recursively, it is find that -th control filter coefficient W is calculated from W, W 2,..., W. So, the first control filter coefficient W can be determined arbitrarily, for example W =, that is the first loudspeaker is through-output. In the result, the control filter coefficients are formulated as follows: where B (m) m W m = = B (m) B (m) m W (2 m M), (9) and A (m) are parameters introduced here and defined as follows: i B (m) = A (m ) i N i= = A (m) i A (m) i A (m) im ( m), (0) A (m) B (m) i B m (m) ( i N, m ), () A (M) i = i d i d ( i N, M). (2) As these formulations show, the control filter coefficients W are calculated from the HTFs between the loudspeakers and the listening positions as well as the HTF d between the virtual sound source and the listener. ISV23, Athens (Greece), 0-4 July 206 3

4 The 23 rd International ongress of Sound and Vibration What described above is with respect to the reproduction of a monaural sound source signal, with attaching binaural characteristics. For reproduction of a binaural sound source signal, a pair of the systems is used, as shown in Fig. 2(b). That is, regarding the binaural sound signal as a pair of monaural sound signals, two systems used for each signals respectively. And furthermore, as shown in Fig. 2(a), on the system for instance for the left channel signal of the binaural sound, its control filter coefficients are calculated under the condition that d = and d = 0, because the sound of the signal have to be delivered only at the left ear of the listener, not at the right. Similarly, on another system, the condition d = 0 and d = is used. These two systems are performed in parallel, and the sounds of the summations of the output signals of both systems are emitted from the loudspeakers. S S Binaural Sound Signal s s d = d = 0 Set d = 0 d = W Set W 2 W (M-) W M W W 2 W (M-) W M E Ob E E Ob E oudspeaker q oudspeaker q 2 oudspeaker q (M-) oudspeaker q M (a) (b) Figure 2: eproduction system configuration for binaural sound source signal. 3. Experiments The dimensions of the loudspeakers locations of the experiment is shown in Fig. 3. In an anechoic chamber, there are two loudspeakers used for the reproduction (so M = 2), located in distance.5 m and in angle ±30 degrees with respect to the center of the listening positions. Another loudspeaker used only for measurements of HTFs d, d of virtual sound sources (not used for reproduction) is located on the circumference of a circle in distance.5 m and in various angles. The dimensions of the listening positions of the experiment is shown in Fig. 4. There are six listening positions lining in a transverse direction with each interval 5 cm. The dummy head Neumann KU00 is used as a head model and stereo microphones. First, the HTFs i, i and d, d are measured (frequency up to 20 khz), next the control filter coefficient W is calculated using the HTFs i, i of only three listening positions (i = 3, 4, 5, so N = 3), as well as d, d. Then, the monaural sound source signal (sound of crow s cry whose main frequency component is in about khz) is convolved with the filter coefficient W, finally the resultant signals are radiated through the loudspeakers. Note that, the listening positions i =, 2, 6 are out of control. Subective evaluations at the center of the listening positions without severe head positioning have resulted in good localization for any direction of the virtual sound source. Obective evaluation have been conducted using interaural cross correlation function (IAF), defined by: IAF(τ) = t2 t P (t)p (t + τ)dt t2 t P 2(t)dt (3) t 2 t P 2(t)dt The left plot of Fig. 5 shows the target IAF of the sound trying to reproduce with the virtual sound source in direction 225 degree.the right plot of Fig. 5 shows the resultant IAFs of the sounds at the six listening positions. Obviously, the IAFs at the listening positions under control are quite close 4 ISV23, Athens (Greece), 0-4 July 206

5 S The 23 rd International ongress of Sound and Vibration eproduction System onfiguration ocations of Virtual Sound Sources Monaural モノラル音源 Sound Source W 0deg deg 35deg θ 90deg d d 270deg d.5m.5m d 35deg 225deg Only for Measurements of HTFs of Virtual Sound Sources 80deg Figure 3: Dimensions of loudspeakers locations of experiment. Monaural Sound Source モノラル音源 W ~ ~ 6 6 実測 HTF ~ ~ 6 6 5cm 任意 N=6 点 d d 目標仮想 Target 音像 Virtual Sound Source S 225deg Figure 4: Dimensions of listening positions of experiment. to that of target, while those out of control are not. The listening area with width of about 0 cm has been achieved using two loudspeakers. 4. eproduction demonstrations Achieving the listening area with width of about 0 cm, the proposed method makes the head positioning easier than in the case of crosstalk cancellation. Therefore, it allows reproduction demonstrations to be held for many and unspecified listeners in such as exhibitions, since it frees them from ISV23, Athens (Greece), 0-4 July 206 5

6 IAF orr coef IAF orr coef - 0 The 23 rd International ongress of Sound and Vibration Target IAF esultant IAF Target 目標 d /d No. 3,2,,4, No. 62 P i / P i [msec] - 0 ( msec ) [msec] - 0 ( msec ) Figure 5: IAFs of the target and resultant sounds, when virtual sound source in direction 225 degree. botheration of strict head positioning. We have already held many reproduction demonstrations under the proposed method using various types of loudspeakers such as; a 4-ch slimmed bar-shaped loudspeaker set below big-screen TV display (width of about 30 cm), a 2-ch smaller loudspeaker set USB-connected to a personal computer (used with distance between loudspeakers of both channels cm), built-in stereo loudspeakers of various devices; a laptop personal computer (width of about 30 cm), a handheld game console (width of about 0 cm), a tablet computer (width of about 25 cm). Any type of loudspeakers used, for calculating adapted control filter coefficients, we measured the HTFs between the loudspeakers and plural shifted listening positions centered at typical listener s head position, in each case. The contents of 3D sound we demonstrated were, for example, dog s bowwow cry sequentially changing its direction of localization clockwise, various sounds localized widthwise or diagonally backward, the sounds are of for instance; opening woody door, baby s laughing voice, pouring beer into glass and drinking it, bell of clock, and various types of voices of speeches (one after the other in various direction like conversation of people), instrumental sound or voice of short speech continuously changing its direction (in other wards, smoothly moving around listener ), binaural recording in situations such as a music live performance and an entertainment park attraction, and so on. These reproduction demonstrations have been performed in not special rooms such as an anechoic chamber but general rooms such as an ordinary meeting room. Many listeners have enoyed the virtual sound effects of the 3D demonstration sounds, we think about percent of listeners have been satisfied to feel good performance of reproduction. 6 ISV23, Athens (Greece), 0-4 July 206

7 The 23 rd International ongress of Sound and Vibration 5. onclusions A new method that achieves wider listening area than traditional crosstalk cancellation is proposed. In this method, filters are formulated for not the binaural sound itself but its interaural complex sound pressure ratio (IA) to be approximately reproduced at plural listening positions. It is assumed that the listener can perceive a virtual sound source in the nearly identical direction by reproducing the IA. Subective and obective experiments have performed to find out that the sound image the listener perceives is almost unchanged even if about 0 centimeters head movement occurs, so the head positioning is easier than that of crosstalk cancellation. The reproduction demonstrations have held using loudspeakers below a TV display, USB loudspeakers connected to a personal computer, and built-in loudspeakers of a personal computer, a handheld game console, and a tablet computer, for many listeners to enoy the virtual sound effects of the binaural sounds. EFEENES. Kyriakakis,., Tsakalides, P. and Holman, T. Surrounded by sound, IEEE Signal Processing Mag., 6 (), 55 66, (999). 2. Atal, B. S. and Schroeder, M.. Simulation of room acoustics using electronic computers, Gravesaner Blatter, 27/28, 24 37, (966). 3. Gardner, W. G. Head tracked 3D audio using loudspeakers, Proc. IEEE Workshop Applicat. Signal Processing Audio Acoust., Mohonk, NY, (997). 4. opez, J. J. and Gonzalez, A. Experimental evaluation of cross-talk cancellation regarding loudspeakers angle of listening, IEEE Signal Processing etters, 8 (), 3 5, (200). ISV23, Athens (Greece), 0-4 July 206 7