Technical University of Denmark

Transcription

1 Technical University of Denmark Masking 1 st semester project Ørsted DTU Acoustic Technology fall 2007 Group 6 Troels Schmidt Lindgreen Kristoffer Ahrens Dickow Reynir Hilmisson Instructor Tobias Piechoviak

2

3 Technical University of Denmark Ørsted DTU Ørsteds Plads Bygning Lyngby Denmark Telephone Title: Masking Course: Acoustic Communication Fall semester 2007 Project group: 6 Participants: Troels Schmidt Lindgreen Kristoffer Ahrens Dickow Reynir Hilmisson Supervisor: Torben Poulsen Instructor: Tobias Piechoviak Date of exercise: November 1 st Synopsis: In this report masking effects are investigated. Masking is when one sound affects another sound, and is something that everyone has experienced. Masking effects can be measured in different ways; backward, simultaneous and forward masking. In this report only simultaneous and forward masking effects are measured, with three test subjects. The results show that the simultaneous masking effect is most significant around the center frequency of the masker, and the forward masking effect decreases as the time delay between the masker and the probe signal grows. Pages: 21 Copies: 4 No part of this report may be published in any form without the consent of the writers.

4

5 Introduction Auditory masking (henceforth masking), is when when the perception of one sound is affected by the presence of another sound. In quiet parts of a song, it is sometimes possible to hear when a singer stamps his foot to keep the pace of the rhythm. It is not possible to hear this effect during a full-on rock anthem because the instruments completely drowns the subdued effect. In other words the stamping is masked by the music. To carry out the analysis of masking, two different experiments are conducted: Simultaneous masking Forward masking When testing simultaneous masking effect, two signal intervals are presented consecutively, separated by a pause. A pure tone is presented together with a narrow band noise in one signal interval while in the other signal interval only the noise is presented. The test subject is asked to indicate which interval carried the pure tone. From the responses of the test subject his/her masking threshold for the given tone and noise can be determined. When measuring forward masking, three signal intervals are presented consecutively, separated by two pauses. All three signal intervals contain noise, but only one of them has a probe signal subsequent. The test subject is asked to indicate which interval contained the probe signal. This time the forward masking threshold is calculated as a function of time delay. Technical University of Denmark, 20 th of November Troels Schmidt Lindgreen Kristoffer Ahrens Dickow Reynir Hilmisson

6

7 Contents 1 Masking theory Masking of pure tones by broadband noise Simultaneous Masking Forward Masking Measuring methods Psychometric functions Measuring setup Simultaneous masking Forward masking Results Measurement results for simultaneous masking Measurement results for forward masking Discussion 17 5 Conclusion 19 Bibliography 21

8 Technical University of Denmark Page 2 Speech intelligibility report

9 October 2007 Chapter 1. Masking theory Chapter 1 Masking theory When exposed to two different sounds (or more), a certain correlation occurs. This correlation is known as masking and is encountered in psychoacoustics. Masking is when the perception of one sound is affected by the presence of another sound. Masking plays a very important role in everyday life. For a conversation outside where it is quiet, little speech power is necessary for the speakers to understand each other. However, if a loud truck passes by, the conversation is severely disturbed. There are two ways of overcoming this phenomenon of masking. Either wait until the truck passes or raise the voice to produce more speech power and greater loudness. The masking threshold is the sound pressure level of a test sound (usually a pure tone), necessary for the test sound to be just audible in the presence of the masker. When masker and test sound are presented, the masking effects can be measured for either simultaneous and not simultaneous masking depending on the presentation. In the latter case, the test sound has to be a short burst or sound impulse which can be presented before, or after, the masker stimulus is turned on. For a stimulus presented before the masker, the masking effect produced is called prestimulus masking or backward masking. This effect is not very strong, but if the stimulus is presented after the masker is turned off, then the masking effect becomes significant, see figure 1.1. Because the test sound is presented after the termination of the masker, the effect is called forward masking or post-stimulus masking. [Zwicker and Fastl, 1999] Page 3

10 Acoustic Communication T. Poulse 4. Masking Chapter 1. Masking theory Technical University of Denmark The term masking is used about the phenomenon that the presence of a given sound (sound A can make another sound (sound B) inaudible. In other words A masks B or B is masked by A. Masking is a very common phenomenon which is experienced almost every day, e.g. when you need to turn down the radio in order to be able to use the telephone. The situation described above is also called simultaneous masking because both the masking signal and the masked signal are present at the same time. This is not the case in backward and forward masking. (Backward and forward refer to time). Simultaneous masking is best describe in the frequency domain and is closely related to the movements of the Basilar membrane in the Figure 1.1: Regions of masking. [Zwicker and Fastl, 1999, p. 78] inner ear. 1.1 Masking of pure tones by broadband noise The masking phenomenon is usually investigated by determining the hearing threshold for a pu tone when various masking signals are present. The threshold determined in this situation is White noise is very called efficient the masked a masker. threshold White contrary noise to the hasabsolute a spectral threshold. density which is independent of frequencies and it produces no pitch or rhythm. In auditory research, the frequency range of white noise is limited to the 20 Hz to 20 khz band. 4.1 Complete masking When using broadband If the ear noise exposed (whiteto noise), white anoise, masking the hearing threshold threshold level is(i.e. found. masked This threshold) will be as threshold is measured shown as in afigure function 4-1 of where the also test the toneabsolute frequency, threshold with different is shown. levels The masked of threshold is show white noise presented. for different levels of the white noise. Figure 4-1. Masking from white noise. Th curves show the masked threshold for different spectrum levels of white noise. From (Zwicker and Fastl 1999) Figure 1.2: Level of test tone just masked by white noise of different density levels (l wn ). The dashed curve indicates the hearing threshold in quiet. From [Zwicker and Fastl, 1999]. The masked thresholds are almost independent of frequency up to about 500 Hz. Above 500 Hz the threshold increases by about 10 db per decade (= 3 db/octave). A 10-dB change in the leve In figure 1.2 the of masking the noise thresholds will also are change shown. the masked Althoughthreshold white noise by 10 has db. a frequencyindependent spectral density, the masking thresholds are straight lines for frequencies below 500 Hz, and If a rise narrow withband increasing signal frequency. is used instead The of thresholds the white rise noise, withthe a masked slope of threshold 3 will be as show db/octave from 500 in Figure Hz. At4-2. lowhere frequencies the masked the masking threshold thresholds is shown lie for about a narrow 17 dbband above signal centred at 250 Hz, 1 khz and 4 khz respectively. Generally, the masking curves have steep slopes (about the given density level. Thus, numbers representing the values of spectral density, l 100 db/octave) towards the low frequency side and less steep slopes wn, (about 60 db/octave) indicate that even negative values of the density level produce masking. Page 4 21 Speech intelligibility report

11 October 2007 Chapter 1. Masking theory When the density level is increased by 10 db, the masking threshold is shifted up by 10 db. This linear behavior of masking is due to broadband white noise. Another thing that characterises Masking the broadband noise is that the frequency dependence of threshold in quiet, almost completely disappear when thresholds masked by broadband noise are measured. [Zwicker and Fastl, 1999] Part 1. Simultaneous masking Pure tones masked by narrowband noise 1.1. Background Instead of measuring masking effects using broadband noise, narrowband noise can be used. A narrowband Simultaneous noise is masking generally is a noise very common with a bandwidth situation equal in which to, orboth smaller the masking than, (masker) and masked signals are presented at the same time. The following figure (Figure 1) shows the critical bandwidth. The critical bandwidth is about 100 Hz for frequencies below 500 masking threshold and the relationship with the absolute hearing threshold. Hz, and 0.2 f for frequencies above 500 Hz [Zwicker and Fastl, 1999]. Figure 1 Level of a test tone that is just masked b critical-band wide noise with a level of 60 db, and centre frequencies of , and 4 khz. The broken curve is the absolute hear threshold in quiet. [3] Figure 1.3: Level of test tone masked by narrowband noise. From [Zwicker and Fastl, 1999, p. 64] Figure 1.3 shows the Narrow threshold band ofnoise pure tones and its masked bandwidth by a critical-band wide noise at center frequencies of 250 Hz, 1 khz and 4 khz. The masking noise has a level of 60 db, and corresponding A bandwidths narrow of noise the noise which 100 is centred Hz, 160at Hz 1000Hz and 700 is Hz, used respectively in this exercise. [Zwicker The bandwidth of thi and Fastl, 1999]. narrow For band a higher noise follows center frequency the 1/3-octave of theband masker, principle: the maximum threshold 6 = f l f c ; u = f c 2 falls as seen on figure 1.3. The difference between the maximum of the masking thresholds f ; Bandwidth = f u f l and the dashed straight line in figure 1.3 is 3 db for the lowest frequency (250 Hz) and where fl, f c and fu are the lower cut-off frequency, centre frequency and the upper cutfrequency. the highest frequency (4 khz). When going from low Measurement frequencies, strategy the masking thresholds show very steep increases, and after reaching maximum threshold a decrease not as steep as the increase is seen. This increase is around In the 100 adaptive db/octave, methods, so in the order presentation to get reliable level data of to the measure signal the depends threshold, the filters previous of these presentation bands havelevels to have and athe very subject step slope answers. of attenuation (more on the course of than 200 db/octave), to exceed the frequency selectivity of our hearing system. Otherwise the The one-up three-down method is a transformed up-down method in which the next presentat frequency response level is of determined the filter is by measured the last and few not responses. that of our This hearing. one-up three-down strategy is based on last three responses and will result in the 79.4% point on the psychometric function. The leve the signal will decrease after three consecutive positive responses Page while 5 the level will increased after one negative response. [2] TAFC means Two Alternative Forced Choice. The test subject shall between two alternative choose the stimulus that contains the signal. The test subject is forced to choose between t alternatives. Figure 2 shows the presentation sequence.

12 Chapter 1. Masking theory Technical University of Denmark 1.2 Simultaneous Masking When two sounds are presented at the same time, a simultaneous masking effect occurs. When measuring simultaneous masking quantitatively, the masking sound is presented Acoustic Communication with limited duration together with short impulses, such as tone bursts or pulses. The signal masked can be shifted in frequency, relative to masker, see figure 1.1 on page 4. another sound (sound B) inaudible. Partial masking is a situation where sound A influences the perception of sound B even though sound B is still audible. The influence is mainly seen in the loudness of sound B. In figure 1.1 an illustration of the masking regions are shown. These regions are backward masking (pre-masking), An example: simultaneous When you masking listen to and a standard forward car-radio masking (post-masking). while you are driving The at, e.g. 100 km/h, figure has two time you will scales, adjust a tthe which level corresponds of the radio to to the a comfortable time relativelevel. to the There onsetwill of the be some background masker and thenoise second from time the scale, engine, t d, the which tires, begins and the where wind thearound maskerthe stops, car (at andleast forward in ordinary cars). Then, masking effectswhen are seen. you [Zwicker come to a and crossing Fastl, or 1999] a traffic light and have to stop you will hear that the radiovolume is much too high. This is an example of partial masking where the background noise masks part of the radio signal and when the background noise disappears the masking disappears too and the radio signal becomes louder than before. (Some modern car radios are equipped with 1.3 Forward a speed Masking dependent automatic level control. The example above is not fully convincing for such cars.) When the masker ends, forward masking occurs. During the time scale given by delay time, t d, found in figure 1.1, the masker is physically inexistent, though it produces masking. This effect can last longer than 100 ms. Backwards masking effect is relative short 4.3 Forward masking T. Poulsen (about 20 ms) compared to forward masking effect, and therefore forward masking is the It has been shown that a strong sound signal can mask another (weak) signal which is presented dominant, non-simultaneous after the strong temporal signal. This masking kind effect. of masking [Zwicker goes and forward Fastl, 1999] in time and is therefore called forward masking. When measuring the forward masking effect, the delay time, t d, is varied. For a high value of this delay Forward time, masking a lower masking is also called threshold post-masking is seen. As(post this = value after). gets Figure smaller, 4-4 the shows the results of a forward masking experiment. The effect lasts for about 200 ms after the end of the strong signal. masking threshold gets higher. Figure 4-4. Threshold of a 20µs Gaussian pressure impulse as a function of the delay, td after the end of a white noise signal. Redrawn from (Zwicker and Fastl 1999) Figure 1.4: Forward masking: peak level of a just-audible 20 µs Gaussian pressure impulse after the end of white noise masker. From [Zwicker and Fastl, 1999, p. 83]. Page 6 Speech intelligibility report 4.4 Backward masking It has been shown that a strong sound signal can mask another (weak) signal which appears before the strong signal. This kind of masking goes back in time and is therefore called backward masking.

13 October 2007 Chapter 1. Masking theory Figure 1.4 shows the masking threshold of a 20 µs Gaussian pressure impulse as a function of the delay time. For a time delay less then 5 ms, no change is in the masking threshold. After 5 ms a small decrease occurs for high levels. The masking threshold gradually decreases until a steady state is reached, around 200 ms. [Zwicker and Fastl, 1999] After describing the theory behind masking, measuring methods will be described in the following chapter. Page 7

14 Chapter 1. Masking theory Technical University of Denmark Page 8 Speech intelligibility report

15 October 2007 Chapter 2. Measuring methods Chapter 2 Measuring methods 2.1 Psychometric functions When doing masking measurements as investigated in this report, results can be expected to be psychometric functions, see figure Correct answers % 70.7 % 50 % Measuring parameter Figure 2.1: Psychometric function. The measuring parameter in this report is the level of test signal measured (pure tone and probe respectively) in db SPL. The dotted lines indicate where the levels corresponding to the 50 %, the 70.7 % and the 79.4 % points can be read from the figure. The results of the simultaneous masking measurements correspond to 79.4 % points on the psychometric function. This means that the thresholds found in the simultaneous masking, describes when the probability that a test subject will give the correct answer is 79.4 %. In the forward masking measurements the results correspond to 70.7 % points on the psychometric function. [Poulsen, 2005b, p. 2+7] Page 9

16 Chapter 2. Measuring methods Technical University of Denmark 2.2 Measuring setup List of equipment Model Description DTU number Sennheiser HDA 200 Headphone Limiter for HDA 200 Computer with soundcard Matlab 2006b Software with measuring procedure Table 2.1: Table of equipment used in the experiment Measuring procedure The measuring setup is shown in figure 2.2. The test subjects are tested using a computer running a pre-programmed Matlab procedure. The sound card of the computer is hooked up to a pair of HDA200 headphones through a limiter to protect the ears of the test subjects. Only the right ear of each test subject is tested in this exercise. The Matlab procedure uses the responses given by the test subjects to decide what the next test signal will be. More detailed descriptions of the actual test procedures are given in the next two sections. Test subject Matlab Limiter Test signal Right ear only Response Figure 2.2: Measuring setup used for both simultaneous- and forward masking. 2.3 Simultaneous masking In this test the masker is a narrow band noise centered at 1 khz. The bandwidth is 1/3 octave, and the upper and lower cut-off frequencies along with the bandwidth are: f u = 1000 Hz = Hz (2.1) f l = 1000 Hz = Hz (2.2) BW = f u f l = Hz (2.3) Page 10 Speech intelligibility report

17 October 2007 Chapter 2. Measuring methods The measuring procedure in Matlab uses a two alternative forced choice procedure and a one-up three-down method. [Poulsen, 2005b]. The test sequence is two noise signals separated by a pause, where one of the signals also carries a pure tone and the other is just noise. The test subject is asked to indicate which of the two signals carries the tone. An example on such a test sequence is given in figure 2.3, where the second signal is the carrier of the pure tone. db SPL Figure 2.3: Presentation sequence of the two alternative forced choice procedure. From [Poulsen, 2005b, p. 3]. 70 Probe signal Noise (masker) Masking The pure tones tested are 500 Hz, 630 Hz, 800 Hz, 1 khz, 1.25 khz, 1.6 khz and 2 khz, and the interval between the two signals in each test sequence is 500 ms. The level of the noise is constant and set to 75 db SPL, see t d figure 2.4. t db SPL 75 Pure tone Noise (masker) Masking f [Hz] Figure 2.4: Narrow band noise centered at 1 khz with pure tone for determining simultaneous masking One-up three-down procedure The one-up three-down method used is a strategy, where (as the name indicates) the level of the test tone is decreased after three consecutive correct responses, but increased after only one incorrect answer and results in a 79.4 % point on the psychometric function. [Poulsen, 2005c, p. 22] How the Matlab procedure works is described in [Poulsen, 2005b, p. 3]. The following is copied directly from [Poulsen, 2005b, p. 3]: The one-up three-down method will be applied to measure the masking threshold. The pure tone will start with a high sound pressure level and then fall down to the threshold level. The step size is varied. A relative larger step size of 4 db will be introduced at the beginning and then decreased gradually down Page 11

18 Chapter 2. Measuring methods Technical University of Denmark to 1 db closer to the threshold. The step size sequence is 4 db, 2 db and 1 db. The change of the step size will take place after a negative response until it reaches the smallest step size. The measurement will stop after 8 reversals and the value of the threshold will be calculated. For more detailed information about starting and using the Matlab procedure, see [Poulsen, 2005b]. 2.4 Forward masking A noise signal is followed by a test signal after a certain time delay. The goal is to determine the masking threshold as a function of the time delay. A one-up two-down method is applied along with a three Alternative Forced Choice (3- AFC) procedure. The 3-AFC procedure implies that the test subject is presented to sequences of three signals seperated by pauses. All three signals contain the test noise, which is 500 ms of Gaussian distribution white noise, but one of them is followed by a 5 ms Gaussian distribution white noise probe signal after a certain (short) delay time. [Poulsen, 2005b, p. 6] See figure 2.5. db SPL 70 Probe signal Noise (masker) Masking t d t Figure 2.5: Noise followed by a probe signal for determining forward masking. 75 db SPL Probe signal The test subject is asked to indicate which of the three signals is subsequent by the probe signal. 500 Noise (masker) Masking The goal is to measure the threshold where the probe signal is just audible to the test subject, at different delay times. The delay times measured are 200 ms, 100 ms, 50 ms, 20 ms, 10 ms and 5 ms. The level of the noise is kept constant at 70 db. The level of the probe also starts at 70 db and varies according to the procedure described below. f [Hz] One-up two-down procedure As when measuring the effect of simultaneous masking the test probe starts at a high sound pressure level, and the tracking size is varied in the sequence 4 db, 4 db, 2 db and 1 db. Page 12 Speech intelligibility report

19 October 2007 Chapter 2. Measuring methods The level of the test probe is decreased after two consecutive correct answers, but increased after only one incorrect answer and results in a 70.7 % point on the psychometric function. [Poulsen, 2005c, p. 22] In the procedure implemented in Matlab, the tests each consists of six trials to determine the threshold at the given time delay. The measurement will stop after eight reversals and Matlab will calculate the threshold of the probe signal at the given time delay. [Poulsen, 2005b, p. 7] For more detailed information about starting and using the Matlab procedure, see [Poulsen, 2005b]. Page 13

20 Chapter 2. Measuring methods Technical University of Denmark Page 14 Speech intelligibility report

21 October 2007 Chapter 3. Results Chapter 3 Results 3.1 Measurement results for simultaneous masking TL KD RH 60 Peak level [db SPL] Frequency test tone [Hz] Figure 3.1: Simultaneous masking plotted on a logarithmic scale. Measurement for test subject RH at 800 Hz is discarded due to error in the result. Test tone [Hz] TL threshold [db SPL ] RH threshold [db SPL ] KD threshold [db SPL ] Table 3.1: Measured results for simultaneous masking. The measurement for RH at 800 Hz is so far from the expected that it will be discarded Page 15

22 Chapter 3. Results Technical University of Denmark as an error. From table 3.1 it is clear that the measurement is wrong. Every other measurement result have 3 significant decimals except the measurement at 800 Hz. 3.2 Measurement results for forward masking TL KD RH 44 Peak level [db SPL] Delay time [ms] Figure 3.2: Forward masking plotted on a linear scale. Delay [ms] TL threshold [db SPL ] RH threshold [db SPL ] KD threshold [db SPL ] Table 3.2: Measured results for forward masking. Page 16 Speech intelligibility report

23 October 2007 Chapter 4. Discussion Chapter 4 Discussion In this discussion the results is compared to theory and discussed. It is assumed that all three test subjects are normal hearing although no preliminary measurements have been conducted to support this assumption. It is also assumed that the equipment used for the measurements in this report is correctly calibrated, since the measurements are based on pre-defined Matlab scripts were the authors have no insight. The measurement results correspond to theory as described in chapter 1. This is easily seen on figure 4.1 where the results for simultaneous masking are plotted on a reference figure from [Poulsen, 2005a, p. 22]. Note that the results can not be directly compared. The reference figure is assumed to be measured on the 79.4 % point of the psychometric function and thereby comparable to the measured results from this report. The results of this report have been measured using 1/3 octave bands where the reference figure uses critical bands as described in chapter 1. This denotes that the measurent results of this report can not be directly compared to the measurent results from [Poulsen, 2005a] but the curve is assumed to be on the same form so the reference figure can be used to support the results in this report. As described in section 3.1 the measurement for simultaneous masking for RH at 800 Hz is so far from the expected that it will be discarded as an error, see figure 3.1. On figure 3.1 it is seen that when going from low to high frequencies, the slope is somewhat steeper before the maximum level is reached compared to the decreasing slope after the maximum. This corresponds to figure 4.1 where the slopes are approximately 75 db/octave in the interval 630 Hz - 1 khz and 40 db/octave in the interval 1 khz - 2 khz. The results for forward masking can be seen on figure 3.2. The figure shows that a stimuli Page 17

24 Chapter 4. Discussion Technical University of Denmark 100 db 80 level of test tone khz frequency of test tone Figure 4.1: Average of measurements for simultaneous masking with a constant noise level of 75 db SPL are plotted on a figure from [Poulsen, 2005a, p. 22]. after a short period is masked more than a stimuli after a long period. The threshold, determining that a person can hear the subsequent stimuli, is very much depending on how long the time delay is. After a 100 ms time delay the threshold is starting to stabilize. Theory shows that after 200 ms the threshold is stable This can not be seen from the results in this report. To show a stable threshold after 200 ms, more tests have to be executed, e.g. at 300 ms and 400 ms time delay. Likewise for thresholds before 5 ms that are also expected to be stable, see figure 1.4 on page 6. The average of the forward masking results correlates to theory, but with deviations. The measurement result for RH at 50 ms time delay deviate from the expected. The reason for this can be from surrounding noise during the measurements. The surrounding noise consisted of people talking and, among others, vibrations into the ear traveling from the floor. The vibrations were so close to the probe in frequency and intensity that is sounded like there was a probe after two signal intervals instead of just one, making it very difficult to detect the right probe. Contrary to the expected test subject KD and RH have a higher threshold at 200 ms than at 100 ms. The deviation is 3 db for both and is considered to be in the normal interval of deviation. The adaptive methods used in this report take minor errors into account by demanding three or two positive answers for the two tests respectively. The test is not stopped before 8 reversals had been carried out, making the influence from isolated errors originated from noise very little. At times the noise were consistent over a period so the surrounding noise could have influenced the measurement. Page 18 Speech intelligibility report

25 October 2007 Chapter 5. Conclusion Chapter 5 Conclusion The scope of this report was to investigate simultaneous masking and forward masking. Theory has been described, mainly considering those two types of masking. Measurements of simultaneous and forward masking were conducted and the setup described. The results were presented and discussed by comparison to theory. From the discussion the following was concluded: The measurement results correspond to theory. The masking threshold for simultaneous masking is worst around the center frequency of the masker. The masking threshold for forward masking becomes better the greater the time delay between the masker and the probe signal. When comparing the results with what could be expected, similar graphs are seen. The masking threshold for simultaneous masking is worst (highest) around the center frequency of the masker. The further away from a masker the threshold becomes significantly better (lower). At lower frequencies than the masker the masking is less effective than at frequencies higher than the masker. The masking threshold, where the test sound is just audible, is lower than the level of the masker at all frequencies within the hearing threshold. The masking threshold for forward masking becomes better the greater the time delay between the masker and the probe signal. At small time delays the masking effect becomes greater making the threshold worse (higher). Page 19

26 This report only measures from 5 ms up to 200 ms and thus it can not be concluded that the threshold will stabilize before 5 ms and after 200 ms. In order to achieve better results, the following propositions are given: Steady and quiet surroundings, perhaps an audiometric booth. A Matlab script that works without constant error messages and faulty measurement.

27 October 2007 Chapter 5. Conclusion Bibliography [Poulsen, 2005a] Poulsen, T. (2005a). Acoustic Communication, Hearing and Speech. 2.0 edition. Only available from DTU campusnet. [Poulsen, 2005b] Poulsen, T. (2005b). Masking. September 11th. edition. Only available from DTU campusnet. [Poulsen, 2005c] Poulsen, T. (2005c). Psychoacoustic Measuring Methods. 2.2 edition. [Zwicker and Fastl, 1999] Zwicker, E. and Fastl, H. (1999). Psychoacoustics. Facts and Models. 2 edition. Available at Ørsted DTU Library Building 352. Page 21