Frequency Response Versus Time-of-Arrival for Typical Cinemas. Louis D. Fielder Dolby Laboratories, Inc

Transcription

1 Frequency Response Versus Time-of-Arrival for Typical Cinemas Louis D. Fielder Dolby Laboratories, Inc

2 Introduction Investigate the suitability of steady state based EQ 1/3 & 1/6 octave steady-state to early-arrival responses early-arrival responses based on first ms of the impulse response divide analyses into front-screen loudspeakers & surround arrays (5.1 ch) comparisons above and below 500 Hz use response averages over 4-16 microphones in audience locations examine a typical 500-seat cinema survey average response characteristics for 18 cinemas 2012 SMPTE e 2012 Annual Technical Conference & Exhibition

3 Basic Definitions and Approach Pink noise stimuli were used. Steady-state responses are simple magnitude-frequency responses. Gated-time responses are the magnitude-frequency responses of: impulse responses a specified time interval from first-arrival sound 4 ms was chosen as the shortest perceptually relevant time interval for front loudspeakers. (based on ear s temporal window and auditory fusion times) 10 ms was chosen as the shortest perceptually relevant time interval for surround arrays. (also includes fact sound arrives from all the individual loudspeakers) gated-time measurements have low frequency limits 4 ms 500 Hz (1/3-octave), 1kHz (1/6-octave) 40 ms 50 Hz (1/3 octave) Frequency responses differences show how closely the above match. a close match means they are equivalent for practical purposes 2012 SMPTE e 2012 Annual Technical Conference & Exhibition

4 Typical 500-Seat Cinema 2012 SMPTE e 2012 Annual Technical Conference & Exhibition

5 5 Steady-State and Early-Arrival Frequency Responses for a Center Loudspeaker in Typical 500-Seat Cinema Level (db) steady state ms 0-40 ms 0-10 ms 0-4 ms X curve Steady-state response is a good match to X-curve for 120 Hz 10 khz Responses are very similar > 500 Hz 0-40 ms response shows a significant drop < 220 Hz -15 1/3-octave smoothing db average across microphones Frequency (Hz) 2012 SMPTE e 2012 Annual Technical Conference & Exhibition

6 Early-Arrival Vs Steady-State Frequency Response Differences for a Center Loudspeaker in Typical 500-Seat Cinema Level Difference (db) steady state 160 ms 4 ms 40 ms 10 ms 1/3-octave smoothing db average across microphones All response differences are very flat with frequency > 500 Hz steady-state response is very similar to any gated-time response > 500 Hz Most of sound energy is direct sound (4 ms = direct sound) 0-40 ms response shows a significant drop below 220 Hz below 220 Hz the steady-state response is very different for the 40 ms gatedtime response Frequency (Hz) 2012 SMPTE e 2012 Annual Technical Conference & Exhibition

7 Early-Arrival Vs, Steady-State Frequency Response Differences for a Left Surround Array in Typical 500-Seat Cinema 0 steady state 160 ms Spectral tilt upward > 500 Hz for gated-time intervals < 160 ms Level Difference (db) ms 40 ms 20 ms 10 ms 4 ms 1.6 db rise between 2 khz and 10 khz small so the steady-state response can be considered very similar Direct sound is a lesser portion of the total than for the center loudspeaker < 300 Hz shows large response differences -10 1/3-octave smoothing db average across microphones Frequency (Hz) 2012 SMPTE e 2012 Annual Technical Conference & Exhibition

8 Average Responses for 18 Cinemas 2012 SMPTE e 2012 Annual Technical Conference & Exhibition

9 Average 0-4 ms Response Differences to Steady State Front Loudspeakers in Small, Medium, and Large Cinemas 0 Level (db) -5 medium large small 1/3-octave smoothing Close match between 4 ms gated time and steady-state responses 500 Hz 16 khz > 500 Hz: steady-state measurements reflect perceptually relevant time intervals Direct to reverberant ratio > 50 % Ignore response roll off > 16 khz due to measurement errors steady-state vs. impulse responses db average across microphones Frequency (Hz) 16k 20k 2012 SMPTE e 2012 Annual Technical Conference & Exhibition

10 0 Level (db) -5 Average 0-10 ms Response Differences to Steady State Surround Arrays in Small, Medium, and Large Cinemas small large medium 1/3-octave smoothing db average across microphones k 20k Modest spectral brightening small = 1.5 db (1-16 khz) medium = 1.8 db (1-16 khz) large = 2.3 db (1-16 khz) > 500 Hz: steady-state responses are reasonable matches to 10 ms Lower direct to reverberant sound ratio (1-5 khz) small -2 db medium -4.5 db large -4 db (atypical? only 4 arrays tested) Ignore response roll off > 16 khz due to measurement errors steady-state vs. impulse responses Frequency (Hz) 2012 SMPTE e 2012 Annual Technical Conference & Exhibition

11 Front Loudspeaker Response Differences 0-4 ms Vs ms from First-Arrival Sounds 5 Level Differences (db) 0-5 1/6-octave smoothing RMS average across microphones 50 front channel loudspeakers in 18 cinemas examined The 4 ms gated-time response is a close match to the steadystate response 47 curves +1.5/-3 db 3 curves with response drop 4 db at 1.1 khz > 1 khz steady-state measurements closely match those of 4 ms Frequency (Hz) 2012 SMPTE e 2012 Annual Technical Conference & Exhibition

12 5 Level Differences (db) 0-5 Surround Array Response Differences 0-10 ms Vs ms from First-Arrival Sounds 1/6-octave smoothing db average offset by -4.5 db RMS average across microphones 28 surround arrays in 18 cinemas examined 10 ms gated-time response a close match to steady-state response 24 curves ± 2.5 db 4 curves with response variations of +4.5/-4 db > 1 khz: steady-state responses are a reasonable match to perceptually relevant gated-time intervals. small spectral upward +1.5 db (2 16 khz) Frequency (Hz) 2012 SMPTE e 2012 Annual Technical Conference & Exhibition

13 0 Level Differences (db) Front Loudspeaker 1/3-Octave Response Differences 0-40 ms Vs Steady State db average across microphones Frequency (Hz) < 500 Hz: the match between gated-time and steady-state measurements becomes poor differences > 11 db below 120 Hz due to room modes with long ring times time scale is >100 ms 40 ms gated-time response is almost always less than the steady-state responses Similar to what is seen for surround arrays (not shown) Steady-state based EQ not optimal but best alternative gated-time based EQ could create higher levels at frequencies with ringing gated-time based EQ bad for organ 2012 SMPTE e 2012 Annual Technical Conference & Exhibition

14 Conclusions The frequency responses of 18 cinemas were surveyed. Cinemas have very high direct to reverberant sound ratios above 500 Hz for front loudspeakers. Surround arrays had less direct sound compared to reverberation No spectral brightening was found for early-arrival sounds compared to the steady state for front loudspeakers and only a small amount for surround arrays. Steady-state based EQ is perceptually relevant > 500 Hz Steady-state based EQ is a good compromise < 500 Hz 2012 SMPTE e 2012 Annual Technical Conference & Exhibition

15 SMPTE Meeting Presentation Frequency Response Versus Time-of-Arrival for Typical Cinemas Louis D. Fielder Dolby Laboratories, Inc, 100 Potrero Ave., San Francisco, CA Written for presentation at the 2012 SMPTE Annual Technical Conference Abstract. Cinema equalization is typically based on the use of 1/3-octave, minimum-phase filtering to adjust the spatial average of the steady-state magnitude response from multiple microphones to the X-curve. This paper explores one aspect of this process, namely whether the use of steady-state response is appropriate. The relationship between early-arrival and steady-state spectral characteristics for typical cinemas was examined. The comparison between early-arrival and steady-state sounds was done via spectral analyses of impulse responses measured at multiple microphone locations within the audience seating area. The cinemas surveyed varied in size between seats and the time-gating intervals varied from 4 ms to that equivalent to steady state. When the survey was done, front loudspeaker measurements above 500 Hz showed little spectral tilt upward toward brightness for early-arrival, compared to steady-state sounds, and a modest upward tilt for surround loudspeaker arrays in the largest cinemas. Larger response differences occurred below 500 Hz. Keywords. Cinema Measurement, Cinema Equalization, Cinema Acoustics, Cinema Standards, Time Frequency Analysis, B-chain The authors are solely responsible for the content of this technical presentation. The technical presentation does not necessarily reflect the official position of the Society of Motion Picture and Television Engineers (SMPTE), and its printing and distribution does not constitute an endorsement of views which may be expressed. This technical presentation is subject to a formal peer-review process by the SMPTE Board of Editors, upon completion of the conference. Citation of this work should state that it is a SMPTE meeting paper. EXAMPLE: Author's Last Name, Initials Title of Presentation, Meeting name and location.: SMPTE. For information about securing permission to reprint or reproduce a technical presentation, please contact SMPTE at jwelch@smpte.org or (3 Barker Ave., White Plains, NY 10601).

16 Introduction Cinema sound systems employ loudspeaker-room equalization to improve the timbre of the sound reproduction. Typically, this equalization process inserts a fractional-octave or parametric filter in the signal path of each channel, applies a pink-noise stimulus, and adjusts the root-mean-sqaure (RMS) average of the 1/3-octave spectra from four or more microphone positions distributed within the audience locations to a target response curve. This target curve is not flat with frequency but employs a significant high-frequency and moderate low-frequency roll off, and is called the X-curve, see SMPTE standard 202M-2010 [1]. The fractional-octave, magnitude-frequency responses of the signals from the measurement microphones are measured when the cinema front loudspeakers and surround arrays are driven sequentially with a pink noise signal. This process measures the entire sound field and is based on the steady-state response. In order to best implement this equalization process, investigators have worked to determine which acoustical measurement method should be used to most accurately determine the perceptually relevant equalization. When the SMPTE standard was first developed in 1977 it was felt that listeners perceived spectral balance based on spectral characteristics of short-duration sounds that were believed to have brighter high-frequency characteristics than those measured by steady-state magnitude-frequency response measurements available at that time, see Allen [2]. Others have investigated the hypothesis that short-duration sounds had a brighter spectral balance than the steady-state sound field, and found it not to be true. Tests showed that early-arrival sounds had a very similar spectral balance as that of the steady-state, magnitude-frequency measurements. One researcher, Holman [3], [4] found similar spectral characteristics between the first 5 and 10 ms first-arrival to steady-state sounds. Another researcher, Munro [5] examined film dubbing and mixing theaters. He found that there was significantly less reverberant sound energy than expected in the sound field, implying that a large portion of a steady-state measurement came from the direct sound and therefore both would have similar spectral characteristics. Others, believing in the mismatch between early-arrival or direct sound compared to steady-state measurements, developed methods in an attempt to assess the perceptually-relevant sound characteristics. Schwenke and Long [6] developed a method that involved the derivation of loudspeaker-room transfer function impulse responses and then varied the analysis time scale as a function of frequency. One difficulty with this approach is that the exact definition of the perceptually significant direct sound as a function of frequency and program material has not been well defined and quantified in the literature. Another approach by Newell, et al. [7], proposed sound measurements at locations outside the audience seating and closer to the loudspeakers to better capture the direct sound characteristics of the loudspeakers. The disadvantage of this approach is that the acoustical mapping between close up measurements and the sound field in the audience area needs to be determined for all the variety of cinema sound configurations. Basic Approach Because of these issues, this study focused on the appropriateness of conventional steady-state measurements made in the audience listening area for the performance of the equalization task. This study looks at the gated-time and steady-state measurements above and below 500 Hz for both front-screen loudspeakers and surround arrays. 2

17 The steady-state response measurement was defined as the spectral magnitude ratio of the measured signal over the stimulus, while the gated-time measurement was defined to be the spectral magnitude-frequency response of a time segment of the derived impulse response extending beyond the first-sound arrival. Because of the intrinsic relationship between time and frequency, shorter time intervals produce measurements accurate only down to a certain frequency as given by equation 1. Minimum valid low frequency Q = 8.65 for 1/6-octave bands, 4.32 for 1/3-octave bands Q f min = 0.5 (1) TimeInterval _ sec onds The 0.5 factor was chosen as a compromise between accuracy and the need to see the lower frequency spectral response values. Some of the response analyses extended the frequency limit slightly lower to 500 (1/3 octaves) or 1000 Hz (1/6 octaves). The perceived frequency characteristics in a room are controlled by the perceived frequency resolution of the ear, called critical bandwidths. This was modeled here by a frequency response smoothing process that was based on the frequency selectivity of the human auditory system as defined by Moore and Glasberg [8] between 100-6,300 Hz. The actual filter response used in the smoothing process was set to zero for attenuations greater than 20 db compared to the Moore-Glasberg function and the frequency characteristics of this filter were then scaled to match either 1/3 or 1/6 octave bandwidths. This filter shape also had the advantage that it closely matched that of a 2 nd -order bandpass filter of equivalent Q factor for attenuations up to 6 db. This filter function was used to provide a rolling average of the FFT squared magnitude values via convolution and the square root of the result is taken to create smoothed fine frequency spaced magnitude values. The 1/3 and 1/6 octave values were chosen because they were similar to what was used in common equalization practice and match the Moore-Glasberg critical-bandwidth estimates, denoted Equivalent Rectangular Bandwidths (ERBs). Figure 1 shows the relationship between Moore s ERBs and 1/3-octave or 1/6-octave bandwidths. The ERB is the most modern and accurate estimate of the frequency resolution of the listener. Figure 1. Critical Bandwidth compared to 1/3 and 1/6 octave bandwidth values 3

18 Examining this figure shows that 1/6 octaves are a good approximation to ERBs for frequencies above 1 khz, while 1/3 octaves are a reasonable approximation for Hz frequency region. Time-synchronous recordings from measurement microphones were made when each channel was driven by a pink noise stimulus of seconds long. These recordings were converted into steady-state or gated-time impulse, magnitude-frequency responses with 1/3 and 1/6 octave smoothing. A sample rate of 48 khz was used. The overall spectral characteristics within the cinema survey were based on the magnitude-frequency averages of the signals from measurement microphones spaced within the audience location, as per SMPTE 202M. The spectral average from multiple microphones was used to limit the effect of localized response variations and room-modal behaviors. This is a very desirable characteristic because it shows general response trends within the listening area, rather than those at a particular location. The use of microphones within the audience listening area was also appropriate because that was where the audience perceives the sound and is mostly likely representative of the listening experience. The microphone signals were either RMS averaged together or averaged together on a decibel basis to assess the characteristics of the entire audience area rather than at only one location. Decibel averaging is used when it is desirable to equally weight each microphones contribution to the response, otherwise it could be argued that only some of the microphone positions dominated the results. For this reason decibel averaging was used in most of the figures. The other averaging method is RMS averaging. The RMS average is typically used in equalization because it has the desirable property of minimizing the effect of spectral notches from just a limited number of the microphones showing up in the response estimate. Since response notches produce equalization peaks, RMS averaging minimizes the number of them. This reduction of peaks is a desirable property because response errors in the form of peaks are generally more audible than notches, see Olive et al. [9] for more details, and Toole and Olive [10] for more details on peak audibility. RMS averaging was used for figures 8 and 9 where the equalization process was modeled more closely. Eighteen typical cinemas, ranging in size from 30 to 1500 seats were surveyed. The measurements spanned 1999 to the present. All measurements were made when the cinemas were configured in 5.1 channel format. Table 1 shows sizes, locations, and number of microphones used for each cinema. Table 1. Characteristics of the cinemas in the study location number of microphones Channels seats Burbank, CA 4 L, C, R, Ls 30 San Francisco, CA 16 L, C, R, Ls, Rs 75 New York, NY 4 L, C, R, Ls, Rs 84 Merced, CA 4 L, C, R, Ls, Rs 110 San Rafael, CA 5 L, C, R, Ls, Rs 130 San Francisco, CA 5 L, C, R, Ls, Rs 140 Merced, CA 4 L, C, R, Ls, Rs 190 San Rafael, CA 5 L, C, R, Ls, Rs 280 San Francisco, CA 4 L, C, R, Ls, Rs 300 New York, NY 4 C, R, Ls, Rs 300 San Francisco, CA 5 L, C, R, Ls, Rs 490 4

19 San Francisco, CA 4 L, C, Ls 500 New York, NY 4 L, C,R 500 Burbank, CA 6 L, C, R, Ls, Rs 500 Los Angeles, CA 4 L, C 1010 New York, NY 4 L, C, R, Ls, Rs 1130 San Francisco, CA 4 L, C, R, Ls 1500 San Francisco, CA 4 C, R, Rs 1500 The cinemas were divided up into three categories: small ( seats), medium ( seats), and large ( seats). For the higher frequency region, a comparison of the spectral characteristics of the steady-state to gated-time impulse response measurements were made and averaged across the audience listening area. The magnitude-frequency characteristics above 500 Hz were used for steady-state and gated-time impulse response measurements covering the first ms from first sound arrival. The time interval of 4 ms was selected as a lower bound for first-arrival sound perception. This was plausible because 4 ms was similar in value to the 5 ms value chosen by Holman [4] in his analysis of transient sounds, where he felt that this interval encompassed the direct-radiation time scale for cinema loudspeakers and was close to the temporal window of the human hearing, as discussed by Plack and Moore [11]. The 4 ms duration was also of similar time scale as the auditory fusion duration, e. g. the time length when frequency rather than time affects control masking, and when direct-arrival sounds and reflected sounds were not separable from each other, see Toole [12], Litovsky et al. [13], and Litovsky and Shinn-Cunningham [14] for more details. Fortunately, the 4 ms time interval was a long enough time period to get accurate measurements of the for 1/3 and 1/6 octave band levels for frequencies above 500 Hz and 1000 Hz, respectively. The lower frequency region was also studied by examining the time-frequency characteristics of cinemas below 500 Hz using the comparison of 40 ms gated impulse responses to steady-state measurements for the Hz frequency region. The values of 40 ms and 50 Hz were chosen as a good compromise between the need to look at lower frequencies and the need to focus on a relatively short time interval. Determination of the Steady-State Response Spectra The steady-state responses were calculated by finding the RMS averaged spectra from each microphone and the RMS averaged stimulus spectrum. The RMS averaged spectra were derived from the magnitude-frequency values from the FFTs of successive blocks of 32,768-samples long, employing a 50 % overlap. The analysis spanned the entire length of the recorded file, minus the last portion, which was smaller than the 50 % of the transform size. Each block was time windowed by a Kaiser window with an alpha = 5, see Harris [15] for details on this window type, called there a Kaiser-Bessel window. Determination of the Gated-Time Response Spectra The gated-impulse response measurements required the calculation of an impulse response and spectral analysis of a time windowed version of that impulse response. The impulse responses for each microphone were first calculated by time aligning the noise signals to the stimulus. Next, the 262,144-sample-length Fast Fourier Transforms (FFTs) on a 5

20 time-windowed version of each successive 262,144-sample block for both were calculated, with a 50 % time overlap employed. The inverse FFT of the frequency band-by-band ratios of average cross-correlation over average autocorrelation values was used to create the impulse response. The cross-correlation was derived from the noise signal and stimulus FFT coefficients, while the autocorrelation was of the stimulus FFT coefficients. A 262,144 sample impulse function resulted. Time windowing was used to improve the background noise immunity by improving the frequency selectivity of the FFTs used. The window was composed of unity gain values in the middle, combined with short fade up/down segments of 256-samples long at each end. The fade up/down shape was based on the first/second halves of a standard Kaiser window of 512 samples, respectively. An alpha factor of 5 was used. The gated-time frequency responses were calculated by selecting a 32,768-sample segment from the longer impulse response, where the 65 th sample was the first time the impulse function reached an absolute value of 30 % of its maximum absolute value. This 30 % start value was chosen to ensure the capture of earlier surround array sound elements in cases where a nearer loudspeaker might have a lower amplitude and be missed. Additionally, it was desirable to allow a short time interval just before the maximum to ensure the capture of the entire transfer function and allow for the action of the fade up portion of the time window without attenuating important parts of the impulse response. The shorted impulse responses were then time windowed by a window defined by the concatenation of the 1 st half of a 128-sample length Kaiser window with an alpha factor of 5, unity-valued samples of a variable sample length, the 2 nd half of the same Kaiser window, and zero valued samples to result in a total length of 32,768 samples. The gated-time interval was defined to be the sum of the variable sample length and 64 samples of the fade-down window. The gated-time interval was set at durations of between ms. Determination of the RMS or Decibel Averaged Spectra When RMS averaging across microphones, the FFT magnitude-frequency values for all the microphones were RMS averaged together, the ratio to the stimulus FFT magnitude-frequency values taken, 1/3 or 1/6 octave smoothing performed, and the result converted to decibel values representing the gated-time, magnitude-frequency characteristic. When decibel averaging across microphones, the ratio of the FFT magnitude-frequency values over the stimulus FFT magnitude-frequency values for each microphone was 1/3 or 1/6 octave smoothed, converted to decibel values, and these were averaged together to represent the magnitude-frequency characteristic. Time-Frequency Characteristics of the Cinemas The analysis of the spectral characteristics of cinemas as a function of gated-time interval is performed by looking at 1/3-octave or 1/6-octave frequency response comparisons for front loudspeakers and surround arrays. Although 1/6-octave smoothing most closely matches the human auditory system above 500 Hz, 1/3-octave smoothing is chosen for the first three analyses below because of the ability to look to a lower frequency, as defined by equation 1. The following will be examined: 1/3-octave gated-time intervals of ms and steady-state responses for a typical 500-seat cinema. 6

21 1/3-octave response differences for gated-time intervals of ms compared to the steady-state responses for the same 500-seat cinema. 1/3-octave response differences above 500 Hz between the 4 ms (front loudspeakers) or 10 ms (surround arrays) gated-time intervals and steady state, averaged together for small, medium, and large cinemas. 1/6-octave response differences between the 4 ms (front loudspeakers) or 10 ms (surround arrays) gated-time intervals and steady state above 1000 Hz, normalized in level over the 1-5 khz frequency region. 1/3-octave response differences below 1000 Hz between the 40 ms gated-time interval and steady state. Time-domain responses at 100 Hz and 150 Hz for four cinema-channel-microphone combinations to better understand low-frequency behavior in cinemas. Examination of the Steady-State and Gated-Time Spectral Characteristics of a Typical 500-seat Cinema The variation of the magnitude frequency response was investigated for a typical cinema of 500 seats in size by finding the db-averaged frequency responses across five microphones for the center channel and left-surround array. Gated-time responses of 4, 10, 20, 40, 80, 160, 320, and 680 ms (not all shown), plus the steady-state response were measured, with the relative levels preserved. Front loudspeaker characteristics Figure 2 shows the response comparison of 4, 10, 40, and 160 ms gated-time responses, along with the steady-state response for the center channel. These values of gated-time intervals were selected from the time intervals tested to demonstrate the important features of the front loudspeaker responses and were still separated sufficiently to create an easy-to-read figure. The level of the steady-state response was adjusted to best match the X-curve for a cinema of this size. The gated-time spectra were all adjusted with the same gain offset to preserve the relative matching of levels compared to the steady-state response. As mentioned earlier, 4 ms was chosen as the lower limit for a perceptually relevant time interval. 7

22 Figure 2. 1/3-octave frequency responses for steady-state and various gated-time intervals for a typical 500-seat cinema front-screen loudspeaker (db averaged) Examination of this figure shows that the steady-state response conforms reasonably closely to the X-curve standard except that the response drops off below 120 Hz and above 10 khz. The response drop off at high frequencies is often seen for the horn-loaded front screen loudspeakers and perforated screen materials typically in use. The responses for the gated-time intervals span the frequency range from the lower frequency limits defined by equation 1 to 20 khz. In this figure, it can be seen that 4 ms gated-time response has the same basic shape of the steady-state response above 540 Hz except for the reduced absolute level. At lower frequencies the close match in shape breaks down, as evidenced by the response drop below 220 Hz when the 40 ms gated-time interval response is examined. Surround array characteristics Figure 3 shows a similar response comparison, but this time for the left-surround array in the same cinema. For display clarity the 4, 10, 20, 40, and 80 ms gated-time responses are shown along with the steady-state one. As before, the level of the steady-state response was adjusted to best match the X-curve for a cinema of this size. The gated-time spectra were all adjusted with the same gain offset to preserve the relative matching of levels compared to the steady-state response. 8

23 Figure 3. 1/3-octave frequency responses for steady-state and various gated-time intervals for a typical 500-seat cinema surround array (db averaged) This time the steady-state response conforms reasonably closely to the X-curve standard except for small response peaks at 90 Hz and 19 khz. More differences are also seen between the shorter gated-time intervals and the steady-state response curve than before. The level of the gated-time responses drops more rapidly because the left-surround array is composed of a number of loudspeakers whose first-arrival wave fronts arrive at different times and cause a slow build up in time. In this case, the first-arrival wave fronts for the array of loudspeakers occur spread over a ms time interval, rather than once. This distributed time sequence of sound arrivals is responsible for significant increase in level as the gated-time interval goes from 4 to 40 ms. The surround loudspeakers also are not designed to keep the ratio of the direct to total sound field ratio as high as in the case of the front loudspeakers, so the earlier arriving sounds are not as large a part of the total sound field. Also different than before, the shortest gated-time responses show a modest increase in spectral brightness compared to the steady-state. Examination of the Spectral Differences between the Steady-State and Gated-Time Spectra of a Typical 500-seat Cinema Although figures 2 and 3 are useful in exploring the differences in frequency response characteristics, a better view of the response trends is available when the gated-time frequency responses are displayed as differences from the steady-state response or its equivalent. As a consequence, this type of display will be used in the remainder of this study. Front loudspeaker characteristics Figure 4 is the alternate representation of the frequency response comparison shown in figure 2. In figure 4 the steady-state response is represented by the heavy line at 0 db and the other 9

24 lines represent the differences between the 4, 10, 40, and 160 ms gated-time and steady-state responses. Figure 4. 1/3-octave frequency response differences between various gated-time intervals and the steady-state response for a typical 500-seat cinema front-screen loudspeaker (db averaged) Examination of this figure shows the difference in the frequency response when gated-time measurements are made. Above 500 Hz there is little difference in the spectral shape except in absolute magnitude, which is a consequence of a uniform frequency build up in sound levels. This similarity above 500 Hz is a strong indication that steady-state measurements can be used to accurately represent the characteristics of all the frequency responses down to gated-time intervals of 4 ms. It is also interesting to note that the sound field at the microphone locations has a majority of the sound energy coming from the direct sound. Below 400 Hz a different picture emerges. In this case the 40 ms gated-time interval shows a substantial build up of sound energy below 200 Hz, that is rather slow and takes ms to happen. This is very different than what is seen above 500 Hz and is likely due to the combination of the reduced ability of the cinema loudspeakers to produce controlled directivity sound radiation patterns and the development of strong low-frequency modes in the cinema space. Steady-state measurements do not give results similar to gated-time intervals in the 40 ms range. Surround array characteristics The characteristics of the example surround array are examined using figure 5, which uses the alternate representation of the frequency response comparison shown in figure 3. Again the steady-state response is represented by the heavy line at 0 db and the other lines represent the differences between the 4, 10, 20, 40, 80, and 160 ms gated-time and steady-state response. 10

25 Figure 5. 1/3-octave frequency response differences between various gated-time intervals and the steady-state response for a typical 500-seat cinema surround array (db averaged) Examination of this figure shows greater differences in the frequency response estimates when gated-time measurements are made. Above 500 Hz there appears to be a spectral tilt upward for gated-time intervals shorter than 160 ms. Additionally, there is a significant drop off in sound level for time intervals less than 40 ms. Investigating this further, it was found that the first-arrival sounds from the loudspeakers in the array arrive distributed in time. As mentioned earlier, the various loudspeaker signals were spread out over a time interval of 35 ms. This makes quantifying the perceptually based estimate of the timbre more complicated because the perception of timbre as a function of time-of-arrival needs to be combined with an understanding of how the listener perceives timbre when many sound arrivals from the same channel arrive within a relatively short time interval. The succeeding analyses will use a compromise value of the 10 ms gated-time interval as the shortest perceptually relevant interval for a typical surround array. When the frequency response difference between the 10 ms gated-time and steady-state response is examined a small spectral tilt upward in frequency is observed. It is 1.6 db between 2 khz and 10 khz. Below 400 Hz the greater build up of sound energy is also evident for the surround array, as it was for the front loudspeaker. The shorter gated-time intervals show a substantial build up of sound energy below 250 Hz. This build up is rather slow and again takes ms to happen. As before, gated-time measurements do not produce the similar spectral results as the steady-state measurements for frequencies below 400 Hz. In summary, in this example cinema the front loudspeaker measurements showed little spectral differences above 500 Hz between the 4 ms gated-time and steady-state measurements, while the surround array had more significant response differences. At low frequencies there was a 11

26 significant build up of sound levels, caused by presence of room modes for both the center loudspeaker and left surround array. Examination of the Average Midrange and High-Frequency Spectral Differences between the Early-Arrival and Steady-State Spectra for Small, Medium, and Large Cinemas The rest of this study investigates whether the 500-seat cinema examined had similar time-frequency characteristics as the other 17 cinemas surveyed. To this end, the average differences between the short gated-time intervals and steady-state responses were investigated for the entire group of 18 cinemas. These were used to assess the accuracy of steady-state frequency response measurements in predicting the frequency response for shorter time intervals from first arrival sound. As before, 1/3-octave smoothing was selected to allow the analysis to extend down to 500 Hz. Decibel averaging across microphones, channels, and cinemas was used to equally weight all contributions in the average. Front loudspeaker characteristics The average time-frequency characteristics of front cinema loudspeakers were examined in figure 6. This shows the decibel averaged difference between the 4 ms gated-time frequency response from the steady-state response for front loudspeakers in three groups of cinemas: seven small cinemas of seats in size, seven medium cinemas of seats in size, and four large cinemas of seats in size. As mentioned earlier, the 4 ms gated-time interval was used because it is unlikely that a shorter time is perceptually relevant for timbre perception. Additionally shown are the response averages for a 680 ms gated-time interval. Since 680 ms is a long time interval compared to significant sound energy arrivals above 500 Hz, it should closely match the steady-state response. 12

27 Figure 6. Average 1/3-octave frequency response differences above 500 Hz between the 4 ms and 680 ms gated-time intervals versus the steady state for small, medium, and large cinema front loudspeakers (db averaged) There are two groups of three curves shown in figure 6. The upper fine lines represent the 680 ms to steady state responses, while the lower three heavy-line curves show the differences between the 4 ms gated-time and steady state responses, all with the correct relative level differences. The solid, dashed, and dotted lines represent responses for small, medium, and large cinemas, respectively. Also shown in figure 6 is a vertical dash dot dot line indicating 16 khz. This will be considered the upper limit for this analysis because cinema equalization typically only corrects the response to this point and it allows ignoring the high frequency mismatch between 680 ms interval impulse response derived and steady-state measurements. Examination of this figure shows that the match of the 680 ms gated-time measurement of the impulse response to the steady-state response is within 0.5 db for the frequency range of ,000 Hz. At frequencies above 16 khz the gated-time averages for the medium and large cinemas show the greatest differences from the steady-state magnitude measurement due to two factors. The first one is an artifact caused by the long averaging time for impulse response derivation when using pink noise measurements lasting seconds in duration. In larger rooms, air currents and small temperature variations cause small variations in the time-of-arrival between the measurement microphone and loudspeaker. This results in a high frequency roll off in the derived impulse response. A second cause for the difference at high frequencies is due to the fact that the steady-state measurement process measures all the acoustic energy and equipment noise in the microphone signal while the impulse-based process used here has a degree of immunity to extraneous noises not related to the stimulus, see Stan et al. [16]. When these differing noise immunities are combined with the rapid high frequency response roll offs seen in some cinema 13

28 loudspeakers, this causes a resultant drop in measurement signal-to-noise ratio and the roll off seen above 16 khz in figure 6 occurs. If the frequency region between ,000 Hz is examined, the average response differences between the 4 ms gated-time and steady-state responses are seen to be quite uniform with frequency for all 3 sizes of cinemas, within a variation of 1-2 db. Additionally, more than 50 % of the sound energy arriving at the audience occurs in the first 4 ms. This is very different than is often believed and indicates that the steady-state measurement is likely to be sufficiently accurate in representing the frequency response over all possible perceptually relevant time intervals for the frequency range of ,000 Hz. Examining the figure shows that the high frequency roll off above 16 khz is the same for the 4 ms and 680 ms gated-time average responses. This indicates that the difference between 4 ms and 680 ms gated-time measurements should match the difference between 4 ms gated-time and steady-state measurements if there were no measurement inaccuracies, allowing the extension of the analysis to 20 khz. As a consequence, the 4 ms to 680 ms response comparison will be used later to represent the early-arrival to steady-state comparison above 500 Hz for figures 8 and 9. Surround array characteristics Next, the average time-frequency characteristics of surround arrays were examined. Figure 7 shows the decibel averaged difference between the 10 ms gated-time frequency response from the steady-state response for the same three groups of cinemas. As mentioned earlier, the 10 ms gated-time interval was used because it was unlikely that a shorter time would be perceptually relevant for timbre perception when the multiple sound wave fronts arrive sequentially from the loudspeakers in a typical surround array. As in the examination of the front channels, 1/3-octave smoothing was used. The response averages for a 680 ms gated-time interval are also shown. Figure 7. Average 1/3-octave frequency response differences above 500 Hz between the 10 ms and 680 ms gated-time intervals versus the steady state for small, medium, and large cinema surround arrays (db averaged) 14

29 Examination of this figure shows that the match of the 680 ms gated-time measurement from the impulse response to the steady-state response is within 0.5 db for the frequency range of ,000 Hz, even better than before. At frequencies above 18 khz the response drop is due to the same causes for the previous example. Again, the vertical dash dot dot line indicates 16 khz. If the frequency region between ,000 Hz is examined, the average response differences between the 10 ms gated-time to steady-state responses show a modest upward slope with frequency. Small cinemas evidence a 1.5 db increase between 1 khz and 16 khz, while medium cinemas a 1.8 db increase, and large cinemas a 2.3 db increase. The proportion of direct sound contribution to the steady-state spectrum is also smaller than for the front loudspeakers. The level difference between the 10 ms gated-time to steady state responses in the 1-5 khz frequency region is approximately -2.2 db for small cinemas, -4.6 db for medium, and -4.1 db for large cinemas. Note that this level decrease is not a monotonic with cinema size, and is likely due to the fact that the estimate was only based on four surround arrays. The drop in early-arrival sound levels compared to the front loudspeaker is expected because of the different goals for front loudspeakers and surround arrays. The front speakers are designed to maximize the amount of early-arrival sound for dialog and audio clarity but the surrounds role is to provide a reverberant environment and minimize the early-arrival sound contribution. Because of the small spectral tilts observed above, the steady-state measurement is slightly less accurate at matching the responses for the shorter time intervals down to 10 ms. However, it is still acceptably accurate considering the response differences expected over the entire audience area. Prediction of Equalization Response Differences above 500 Hz between Early-Arrival and Steady-State Measurements The previous section examined the average 1/3-octave frequency response differences between 4 ms or 10 ms gated-time intervals and steady-state responses. This time a 1/6-octave response comparison was made. The 680 ms gated-time response was used instead of the steady-state response because 680 ms time interval was long enough to capture any significant reverberant sound energy above 500 Hz and avoided the small measurement inaccuracies between impulse-response-based and steady-state measurements. This analysis reinvestigated the suitability of steady-state measurements to give perceptually relevant results by more accurately modeling an equalization process. This time it was based on 1/6 octaves, which were used to more accurately model the auditory bandwidth of the ear above 1 khz. The 1/6-ocatve response differences are shown for individual front loudspeakers or surround arrays, rather than as averages to highlight any possible differences. Since equalization tasks typically use RMS averaging across measurement microphones, this is used here. Additionally, the differences are normalized in level to match 0 db over the range of 1-10 khz to allow the focus to be only on the response variations as function of frequency. Front loudspeaker characteristics Figure 8 shows the normalized response differences between the 4 ms and 680 ms gated-time responses. The 50 front loudspeakers found in the 18 cinemas surveyed are shown as individual lines. The frequency range extends 1-20 khz. 15

30 Figure 8. 1/6-octave frequency response differences between the first 4 ms and 680 ms from first-arrival sounds for 50 front-screen channels (RMS averaged) Examination of figure 8 shows that all but three front loudspeakers possess response differences between 4 ms and 680 ms within the range of +1.5/-3 db, while the three front loudspeakers from one small cinema show a short term response drop off of db at 1100 Hz. Because 94 % of the examples show small response differences between gated-time intervals of 4 ms and 680 ms, steady-state measurements accurately represent perceived timbre above 1 khz, supporting the results from the analysis of average response differences in figure 6. Surround array characteristics Figure 9 shows the normalized response differences between the 10 ms and 680 ms gated-time responses. The 28 surround arrays found in the 18 cinemas surveyed are shown as individual lines and the frequency range extends 1-20 khz. 16

31 Figure 9. 1/6-octave frequency response differences between the first 10 ms and 683 ms from first-arrival sounds for 28 surround-array channels (RMS averaged) Examination of figure 9 shows larger variations than those seen for the front loudspeakers. This time they vary over a +4.5 to -4 db range. If 24 of the 28 surround arrays are considered the response variation reduces to ± 2.5 db. One significant change from the front loudspeaker characteristics shown earlier is the trend towards a spectral tilt upward in frequency. This can be seen by examination of the average response difference, shown in the figure with an offset -4.5 db for clarity and shown as the heavy dashed line. It shows a 1 db increase at 10 khz and a 1.5 db difference at 16 khz, relative to 2 khz, similar to the db-average-based results of figure 7. The conclusions to be drawn from figure 9 are similar to those from the averages in figure 7. Steady-state responses or their equivalent are reasonably accurate in most circumstances for predicting the surround array responses down to the 10 ms gated-time interval considered. Examination of the Average Low-Frequency Spectral Differences between the First 40 ms and Steady-State Spectra The lower frequency response characteristics of cinemas were investigated by looking at the decibel averaged responses across the measurement microphones for the loudspeakers and arrays in the 18 cinemas. The individual loudspeaker and surround array frequency response differences for each loudspeaker or array between 40 ms gated-time and steady-state values were evaluated at frequencies between Hz, with 1/3-octave smoothing. As before the results were divided into front loudspeakers and surround arrays. LFE channels were not studied. Front loudspeaker characteristics Figure 10 shows the 40 ms gated-time to steady-state response differences for the 50 front loudspeakers in the 18 cinemas surveyed. The frequency range shown is Hz but the 17

32 focus is on the Hz region. At these frequencies 1/3-octave smoothing is a good match to the auditory bandwidth of the ear. Figure 10. 1/3-octave frequency response differences between the first 40 ms and steady-state sounds for 50 front-screen channels (db averaged) Examination of figure 10 shows that the response differences are almost always negative because the sound level energy increases as the sound field is sampled over longer time intervals. The small positive values observed are either due to small measurement errors, or in the case of the small positive peaks at 600 and 750 Hz, due to destructive interference for long time intervals. The negative differences are due to sound level energy build ups for time intervals greater than the 40 ms time interval. These vary tremendously from loudspeaker to loudspeaker. Some loudspeaker-cinema combinations show very small level increases, while others have level changes of more than 10 db. The differences become significant below 500 Hz, particularly so below 200 Hz. Surround array characteristics Figure 11 shows the 40 ms gated-time to steady-state response differences for the 28 surround arrays in the 18 cinemas surveyed. Again the frequency range shown is Hz with a focus on Hz. As before, 1/3-octave smoothing is used. 18

33 Figure 11. 1/3-octave frequency response differences between the first 40 ms and steady-state sounds for 28 surround-array channels (db averaged) Figure 11 shows a similar situation for the response differences for the surround arrays. This time there are no small positive frequency-response differences and generally these negative spectral differences were less extreme than for the case of the front loudspeakers. This is likely a result of the fact that the distributed nature of the sound radiation from multiple loudspeakers does not as effectively excite the room modes at low frequencies. This time the significant differences start below 400 Hz and tend to be spread out more in frequency. The curve with the most extreme differences is from a large 1500-seat cinema and shows evidence of significant and broad response anomalies below 300 Hz. The presence of significant response differences between the 40 ms gated-time and steady-state responses demonstrate that steady-state response measurements give significantly different results than measurements looking over a more limited time scale. Examination of the Low-Frequency Time-Domain Characteristics of Example Cinemas with Slow Build-up of Sound Levels The nature of the sound propagation at low frequencies was explored further by the use of a number of examples of cinema-loudspeaker-microphone combinations shown in the previous section. These were chosen to represent the wide range of response possibilities and shown in figure 12. Examples from small, medium, and large cinemas were chosen. 19