The acoustics of Roman Odeion of Patras: comparing simulations and acoustic measurements

Transcription

1 The acoustics of Roman Odeion of Patras: comparing simulations and acoustic measurements Stamatis Vassilantonopoulos Electrical & Computer Engineering Dept., University of Patras, 265 Patras, Greece, Panagiotis Hatziantoniou, John Worley, John Mourjopoulos Audiogroup, Wire Communications Laboratory, Electrical & Computer Engineering Dept.,University of Patras, 265 Patras, Greece, {hagianto, worley, Juha Merimaa Laboratory of Acoustics and Audio Signal Processing, Helsinki University of Technology, P. O. Box 3, 215 TKK, Finland, The acoustics of Patras Odeion, a well-preserved open-air ancient Roman theatre, have been investigated via computer model simulation and a novel set of acoustic measurements. These tests were aiming at verifying the accuracy of the modelling procedure, optimising its parameters, and investigating the various means for describing the 3-D sound field for acoustic analysis and auralization. The detailed computer simulation of the theatre and acoustic measurements were made for 5 identical listener / receiver positions, respectively. The measurements consisted of single-channel impulse responses, 8- channel microphone array responses and 2-channel binaural responses. From the analysis of the singlechannel response measurements and the simulation results, comparisons are drawn related to various acoustic parameters (RASTI, D-5, C-8, etc). The microphone array responses were processed to generate detailed spatial results for the theatre s sound field. Finally, the binaural response measurements were employed for comparisons via psychoacoustical tests, between computer simulation-generated auralizations and binaural audio signals. 1 Introduction In the past years, the use of advanced acoustic modeling tools has allowed for comprehensive studies of the acoustics of many historic monuments [1], [2]. In addition, over the recent years, response measurements have evolved so that directional analysis of the sound field can be performed [4], [5]. Aim of this work is to compare in detail alternative measurement and acoustic simulation methods, with respect to an ancient open-air theatre. The Patras s Odeion is a well-preserved Roman amphitheater, located within the modern city of Patras, in southern Greece. The theater was constructed early in the 2 nd century AD, having a capacity of 23 spectators, but became disused after the 3 rd century AD. The theater was discovered in 1889 covered by earth and was later restored, used until this day for music and theatrical performances. A special feature of the theater is that brick was employed for the construction of the concave structure ( koilon in ancient Greek or cavea in Latin) and its façade (although seating and aisles were covered by marble) and its complex and large brick stage structure ( skene or scaena ), typical of Roman theaters, which is still largely preserved today. Some properties of the theater are listed in Table 1. Table 1. Properties of Patras Roman Odeon Radius (m) Slope ( ) Rows Capacity Figure 1: plan-view with noted source (S) and receiver (R) positions. 2 Methods 2.1 Acoustic measurements The measurements were performed during July 24, using laptop computer with an RME soundcard and an ATC SMC 2-2 Active Monitor as a sound source. Measurements were obtained from positions identical to those used for simulations, as shown in Figure 1, see also [1]. The source was placed at a height of 1,5 m in the center of the scene (S1) and also at a position 5m closer to the koilon (S2), see Figure 2. The measurements were performed using a sinusoidal sweep excitation. The logarithmic sweep was 2197

2 generated and deconvolved in the frequency domain, as described in [3]. The level of the excitation signal was set manually to yield an optimal signal-to-noise ratio, which was in average around 28 db. The measurements consisted of single-channel impulse responses, 8-channel microphone array responses and 2-channel binaural responses (see Figure 3). The single-channel responses were measured with an omnidirectional measurement microphone. The 8- channel measurements were conducted with a custom 3-D array of microphones, allowing sound intensity analysis at each Cartesian coordinate axis using standard pairwise measurement methods [4,5]. Therefore, the directional properties of the theatre s sound field could be analysed and compared to the corresponding computer model simulations. As it is known, the importance of early reflections, especially side reflections generated via the theatre s concave structure, can contribute to the unique spatial impression generated to listeners in such spaces, an acoustic feature that is not easily measured via the traditional acoustic tests. For the binaural measurements, 2 miniature microphones were inserted to the ears of a listener seated at each measuring position. Figure 4. For the simulations, 14 typical listening positions R 1 to R 14 and two source positions S 1 and S 2 in the circular stage (orchestra) were chosen. Alternative options were examined, one set including the background stage ( proskenion ) and a set without this structure. For each case, and for the frequency range of Hz a number of established acoustic parameters were evaluated, such as RASTI, T S, D-5, C-8, LEF, G-1, and SPL. Figure 3: Equipment and measurement set-up Figure 2: Test set-up with source at the centre of the stage 2.2 Acoustic Modelling For the computer model simulation of the theatre, detailed architectural drawings were obtained with the help of the 6 th Office of Classical and Prehistoric Antiquities, Patras and the 2 nd Revenue for Later Monuments [1]. Data were also obtained form various historical records, indicating details of the early use and geometrical features of the theater. The electronic model was then generated in AutoCad (v.14) and the geometrical coordinates were transferred into a commercial acoustic simulation software (Catt- Acoustic, v.7.2b), producing the model shown in Figure 4: The theater s 3-D computer model 3 Results 3.1 Time domain A comparison of measured and simulated theatre impulse responses (evaluated here as echogram Energy level) is presented in Figures 5 and 6. It can be observed that the model describes accurately the earliest reflections (5-1 ms after the direct signal), arriving from on the stage floor. In most cases, secondary side reflections around 4 ms, were also properly simulated. There are also some differences between simulation and measurement. 2198

3 -2 s1r s1r s1r s1r Figure 5: impulse responses (source: S1, Receiver positions from bottom to top: R2, R4, R8, R12). Vertical axis: Energy Level (db), horizontal axis: Time (ms). These are due to the simplified use of diffraction in the model and also the existence of high levels of background noise in the measurements. As will become later evident, the geometric details of the theater s surfaces (seating aisles edges, etc.), contribute to some significant amount of diffracted energy which cannot be described by the model. Furthermore, noise from the modern-day city, contributes a noise floor level which is apparent in the measured responses and not in the simulated ones. Figure 6: impulse responses (source: S1, Receiver positions from bottom to top: R2, R4, R8, R12). Vertical axis: Energy Level (db), horizontal axis: Time (ms). 3.2 Spatial domain The analysis of the 8-channel microphone signals, arranged as a 3-D array yields directional information of the impulse response data, which can be related to the corresponding angle of arrival of each discrete reflection [4,5]. For each 8-channel response, 2 timefrequency analysis maps were 2199

4 active intensity vectors plotted on top of an omnidirectional spectrogram. The arrows thus indicate the net flow of sound energy within each time frequency bin, and they can be used to estimate the direction of arrival of discrete sound events. The spectrogram and the lengths of the arrows are presented on a logarithmic (db) scale, and the data have been thresholded to 35 db in order to hide smallest details and the background noise. Careful observation of the figures confirms that the first early reflections (approx. at 5 ms after the direct sound) arrive from the stage floor (see the direction of the arrows in the median-plane plots). The next set of reflections (approx. at 1 ms after the direct sound), arrives from behind the listener, due to the reflections at the vertical sections of the aisles. Some diffuse lowlevel sound (reflections and diffraction) appears to arrive at early times from random directions, but significant broadening of the reflection angle of arrival only appears beyond 3 ms after the direct signal. When the receiver is moved to off-centre positions (e.g. R8), then, the reflections arrive from wider angles, indicating higher level of contribution from lateral reflection energy. 3.3 Frequency domain Magnitude (db), 2 db/div s1r12 s1r8 s1r4 s1r2 s1r1 1 1k 1k log Frequency (Hz) Figure 8: 1/3 octave magnitude frequency response of the theater at different receiver positions Figure 7: Directional time-frequency plots of the measured 8-ch responses (source: S1, Receiver positions from bottom to top: R2, R4, R8, R12). computed, corresponding to the horizontal plane and to the median plane. The panels in Figure 7 consist of It is also useful to consider the 1/3 octave frequency response (magnitude), derived from the 1-channel microphone data (Figure 8). It can be observed that the theatre responds mostly to the 3 Hz 6 khz frequency range, having a significant attenuation around 2 Hz. This feature seemed to be largely independent of the listening position, having also been measured for the ancient theater of Epidaurus [6], and can be related to the effect of the reflection from the stage floor. 22

5 3.4 Acoustic parameter comparison These results were obtained from processing the measured responses (1-channel) and the computer model simulations, to derive the well-established sets of acoustic parameters (C8, D5, RASTI) and are presented as functions of source / receiver distance in the Figure 9, for the centre-positions and for different source positions (S1 and S2). It can be observed that measurements and predictions of the clarity criterion (C8) are in good agreement. As expected, the clarity is reduced with distance and improved slightly for the source moving closer to the audience (S2). C8 (db) D5 (%) RASTI (%) source 1: θ=5 source 2: Source/Receiver Distance (m) source 1: θ=5 source 2: Source/Receiver Distance (m) source 1: θ=5 source 2: Source/Receiver Distance (m) (a) (b) Figure 9: measured and simulated results for (a) clarity (C8), (b) definition (D5), (c) speech intelligibility results (RASTI) as function of source receiver distance and source position. (c) For the definition criterion (D5), the simulations appear to underestimate slightly the result. Speech intelligibility (as evaluated via the RASTI) seems to be very high in all positions. Significantly, measured responses produced higher RASTI results than simulations (especially for the central stage source position). Previous studies have also confirmed the underestimation from computer models of some acoustical parameters in open-air theatres [2,6]. 4 Comparison between simulation and actual auralization The purpose of perceptual evaluation is firstly, to assess the perceptual accuracy of virtual acoustic reconstruction, and secondly, considering the similarity between the models and the actual theatres physical measurements, assess the perceptibility of these small differences. The tests were conducted with ten participants aged between 18 and 35 years old, native Greek speakers. All of the participants reported normal hearing. The majority of the participants (8/1) were inexperienced in psychoacoustic tests. The stimuli employed were recordings of ancient Greek speech (close-microphone, 44.1 khz sampling rate, 1.5 s duration). The speech utterance was convolved with two pairs of the Binaural Room Impulse Responses (BRIRs), one being the measured response derived in the actual theatre, the other derived from the computer model/simulation for the same position. Each stimulus contained three utterances, firstly the reference, then followed by two further utterances, one of which was the same as the reference. The reference utterance was followed by 1 s of silence, and then the listener heard the two further utterances, with a 5 ms inter-stimulus-interval (ISI). One of the second pair of utterances was the same as the reference, and the other was the opposite utterance. For instance, an A_AB pair had the modeled utterance as the reference, which after 1 s of silence was followed by the same modeled utterance, then after 5 ms of silence the listener heard the utterance convolved with the measured BRIR. The total duration of each trail was 6 s. The overall levels of the convolved utterances were matched by ear. The stimuli were presented from a desktop personal computer, to a pair of closed back headphones (Ultrasone ProLine 55). The experiment took the form of a 2I-2AFC ABX task, whereby the listener was presented with a reference stimulus, and had to judge which of two succeeding stimuli was different from the reference. Thus, on each trial the listener heard three speech utterances, only one of which was different. The stimuli were presented and the participants responses recorded. Once the listeners made their response, the next stimulus was 221

6 immediately presented in a random order. In total each listener was presented with 4 trials, consisting of 4 possible utterance combinations (A_AB; A_BA; B_AB; B_BA), with 1 repetitions of each. The collected psychoacoustical data shows that in most cases listeners identified the mismatched stimulus. The listeners are correctly selecting the mismatched stimuli on 85.5% (standard error of the mean = 2.73%) of trials when the reference is from the simulated BRIRs, and on 79.5% (standard error of the mean = 4.44%) of trials when the reference uses the measured BRIRs. With chance performance at 5%, the participants are significantly identifying the mismatched stimuli more often than would be expected by chance alone, irrespective of the reference stimuli. Performance at identifying the mismatched stimuli is shown to be statistically significantly above chance by t-tests ( -.t(9) = , p <.1; t(9) = 6.647, p <.1). The ANOVA showed a non significant effect of which utterance served as the reference stimulus F(1,9) = 5.6, p = Conclusions The theatre simulations appeared to have an acceptable degree of similarity to the measured results. Nevertheless, some acoustic parameters were underestimated by the model, probably due to the simplified simulation of diffusion which appears to play a significant role to the open-air ancient theatre soundfield. Auralizations from the model and measurements seemed to be easily distinguished by listeners, but this may be partly attributed to mismatch between HRTFs, in addition to imperfect description of the absorption and diffusion of the model's surfaces. The significance of the extraction of directional information from measured and simulated responses was also confirmed, allowing for a better interpretation of the time, frequency and directional soundfield information in such theatres. Acknowledgement The work of John Worley and Juha Merimaa has been supported by the research training network for Hearing Organisation and Recognition of Speech in Europe (HOARSE, HPRN-CT ). References Figure 1: mean perceived differences as a function of reference stimuli type (standard errors over 1 subjects). The main finding from this study is that irrespective of the reference stimuli, listeners can easily tell the difference between the simulated and measured BRIRs. This result must be partially attributed to the difference in the HRTFs employed for measurements and simulations, which are not easily compensated via preprocessing. Also, this result can be due to small mismatch between the simulated and the actual response, due to inaccuracies in the modeled absorption properties of the theater surfaces, especially the stage structure. It must be noted that after some alterations of the model s absorption for this surface (stage), more accurate BRIR simulations were derived, as could be verified by informal listening tests which were not subjected to the above formal evaluation procedure. [1] S. L. Vassilantonopoulos, J. N. Mourjopoulos: A Study of Ancient Greek and Roman Theater Acoustics, Acoustica 89, 22. [2] A.Gade, M.Lisa, C.Lynge, J. Holger Rindel, "Roman Theatres Comparison of acoustic measurements and simulation results from Aspendos Theatre, Turkey", Proceedings of the 18th International Congress on Acoustics, Kyoto, Japan 24. [3] S. Müller and P. Massarani, Transfer-function measurement with sweeps, J. Audio Eng. Soc. 49(6), , 21. [4] J. Merimaa, T. Lokki, T. Peltonen, and M. Karjalainen, Measurement, analysis, and visualization of directional room responses, AES 111th Convention, New York, NY, USA, 21. Preprint [5] J. Merimaa, Applications of a 3-D microphone array, AES 112th Convention, Munich, Germany, 22. Preprint 551. [6] Vassilantonopoulos et. al. Measurements and Analysis of the Acoustics of the Ancient Theater of Epidauros, 2nd Conference of the Hellenic Institute of Acoustics, 24, Thessaloniki, Greece, (in Greek). 222