Acoustic absorption and reflection as train interior design elements Nils Maximilian Zapka

Transcription

1 Acoustic absorption and reflection as train interior design elements Nils Maximilian Zapka Master s Degree Project TRITA-AVE 2013:31 ISSN

2

3 Abstract Acoustic comfort in train compartments is discussed in relation to speech intelligibility. Passive acoustic design solutions were developed, resulting in two screen prototypes that use absorption and reflection to lower the level of speech intelligibility. A method of evaluating these prototype s effect on speech intelligibility was developed based on Speech Intelligibility Index (SII) measurements. Measurements were carried out in a train compartment mockup that resembled a 3 m section of a train coach. A computer ray tracing model of the measurement setup was created in the ODEON ray tracing software and verified against the mockup. Simulations were carried out and compared to the measurement results and the applicability of the ray tracing model discussed. The measurements and subsequent evaluation according to the method developed here were able to predict the prototype s effect on speech intelligibility and are found to be applicable to other future setups. Acknowledgments The author would like to thank his supervisors Leping Feng at KTH and Adam Mirza and Siv Leth at Bombardier Transportation in Västerås. The guidance and feedback throughout the process of this work was essential for its completion. Also the feedback given on the presentations during the last months by the staff of the acoustics department at Bombardier Transportation influenced the direction the work followed. Thanks goes also to Claes Kastby, who simultaneously worked on his Master thesis on active noise control for Bombardier Transportation and shared an office with the author for months, thereby being first in line for discussing issues the author faced. iii

4 Contents 1 Introduction Task description Theory Objective measures of speech intelligibility Simulation Geometrical room acoustics ODEON Ray tracing software Using screens Screens in room acoustics Dealing with diffraction Method The train compartment model The mockup The ray tracing model The artificial human head The screen prototypes Measuring SII Measurements and simulations Measurements on the acoustically shielding box Results ODEON simulations Conclusions 30 6 Discussion Measurements Simulations

5 1 Introduction A major concern of railway passenger transportation companies is the improvement of efficiency, both concerning energy consumption and the number of passengers traveling. In the competition between railway travel and travel by airplane and car, a point of advantage of trains that often is stressed by the railway industry is the higher level of passenger comfort. The term passenger comfort is diverse, though, as it involves various aspects such as thermal, luminous, spacious and acoustic comfort. Acoustic comfort in railway transportation is in itself a vast field of study, as the sources of acoustic disturbance are as numerous as they are different, e.g. rolling and wind noise, noisy machinery, vibrating floors, walls, ceilings, doors, panels etc. These sources of disturbance can be handled to some degree by treating their mechanism of excitation, yet there will still be some sound field surrounding the passenger in a train. Further sources of noise, e.g. other passengers talking to each other or into cell phones cannot be treated in the same way. Thus, the sound field the passengers are exposed to can be regarded as the sum of train/cabin noise and speech noise. This work will only treat measures of acoustic comfort enhancement that are taken close to the passenger, i.e. accepting the train noise and speech noise as they are and introduce passive measures close to the passenger and examine which criteria these measures need to fulfill. Since acoustic comfort is a subjective perception, care must be taken when trying to assess the influence of measures to treat either train noise, speech noise or both. Kuwano et al. [1] have described this viewpoint on the issue of acoustic comfort in high speed trains and the interaction of cabin noise and conversation regarding annoyance. They found that lower noise levels not necessarily lead to less annoyance, since a conversation can be more of a disturbance when the level of cabin noise is low than if it were high. Thus, cabin noise can if it in itself is not annoying serve as pleasant masking noise, even if it raises the total sound level around the passenger. Similar results are shown by Venetjoki et al. [2], who evaluated the effect of speech intelligibility on task performance. Their study was conducted in the context of open plan offices, which can be regarded as an environment similar to a train compartment, especially since one of the advantages of train travel is that the time spent on the train can be used for work. The study shows that the performance of demanding tasks such as proofreading or verbal tasks deteriorates with growing levels of speech intelligibility. This work was carried out in a laboratory environment and aimed at formulating a method to evaluate passive shielding solutions by making use of acoustic quantities. This method was the basis of a series of measurements and simulations and the results were analyzed with regard to acoustic comfort, especially speech intelligibility. 1

6 1.1 Task description A mockup model of a section of a train cabin had been designed and built and was ready to be used when this thesis work commenced. A ray tracing computer model was to be created with the ODEON ray tracing software and this computer model was in turn to be verified with respect to similarity to the mockup. One goal of the work was to find possible prototypes of solutions to use screening constructions to enhance acoustic comfort, which is a subjective perception. A method needed to be found that can quantify the performance of these constructions regarding acoustic comfort by using acoustic quantities. Some prototypes were to be built and measurements and ray tracing simulations to be performed. The measurements and simulations were to be used as a basis to draw conclusions about the prototypes performance and generate insight into how the process towards interior design elements that focus on the alteration of speech intelligibility should be planned and evaluated. 2

7 2 Theory 2.1 Objective measures of speech intelligibility Quantifying speech intelligibility was historically done through subjective assessments. Since the 1960 s a number of objective measures have been proposed which use acoustic quantities to describe the subjective perception of speech intelligibility. Below an overview of the most widely used measures is given. Note that these are not global room parameters, but measures to describe source receiver signal transmission. Furthermore should be noted that they are measured indices and were not developed to straightforward predict speech intelligibility. Articulation index One of the first methods to assess speech intelligibility objectively by means of acoustic measurements was the Articulation Index (AI). Its purpose is to quantify speech intelligibility as a single number value between 0 and 1 where 0 denotes no intelligibility and 1 denotes full intelligibility. It is in its essence a signal-to-noise ratio. Distortions are only accounted for in the frequency domain, e.g. interfering noise or band-pass filtering. Time domain effects like reverberation are not considered, thus AI is no good measure in highly reverberant sound fields. Still, it is regarded as a good measure in non-reverberating enclosures, e.g. aircraft cabins [3]. AI can also be used inversely to describe low intelligibility, then mostly denoted as high Privacy Index (PI) a quality measure of open plan office landscapes etc. Note that PI nowadays refers to the privacy index measured from the SII definition shown below. Speech Transmission Index The Speech Transmission Index (STI) uses modulation depth as an acoustic quantity to describe speech intelligibility. It makes use of the nature of human speech as carrier frequencies and low-frequency modulations. The underlying rationale of the characterization of the speech transmission path by modulation depth was described by Houtgast and Steeneken [4, 5]. They introduced the concept of the Modulation Transfer Function (MTF) originally a performance measure of optical systems into the field of room acoustics. Since a speech signal can be described as an intensity pattern distributed over frequency and time, its alteration or the damage done to it by the transmission within an enclosure can be quantified by the smearing of the temporal intensity distribution. In detail this damage originates from the great number of transmission paths with varying time delays. Additionally the signal-to-noise ratio at the receiver s end is corrupted by interfering noise, see Figure 1. The MTF shown describes the effect as the fully modulated signal reaching the receiver as I in = I 0 (1 + cos 2πf m t) (1) I out = I noise + I 0 (1 + m(f ) cos 2πf m (t + τ)), (2) 3

8 where the degree of modulation has decreased from unity to m(f ), which is the MTF, a function of modulation frequency F. Figure 1: The description of the speech transmission path with MTF m(f ) [5]. When calculating STI the relations between the MTF and the two acoustic parameters reverberation and signal to noise ratio need to be determined. For reverberation this relation is two dimensional, since the MTF is a function of both the modulation frequency and the carrier frequency band. In the signal to noise ratio case the MTF is independent from modulation frequency, as seen in (2). A complete STI measurement involves comparing the modulation depth of a signal sent through the tested transmission path in the seven octave bands which are part of the human speech range. In each octave band fourteen modulation frequencies are used, resulting in 98 data points. STI is subject to continuous improvement and depending on which version is used (e.g. original STI or revised STI (ST I r ), introduced in 2003) different influences on the transmitted signal are accounted for: Non contiguous frequency transfer, band pass filtering, auditory masking or the absolute hearing threshold. Thus, the version applied needs to be stated when performing measurements. To make assessment procedures more convenient the Rapid Speech Transmission Index (RaSTI), later called Room Acoustic Speech Transmission Index (RASTI), was developed. Its difference compared to the full STI measurement is that only nine modulation frequencies are used spread over two octave bands, reducing the number of data points to less than a tenth. The octave bands tested are the 500 Hz (vowels) and 2000 Hz (consonants) bands. One drawback of this simplification is that the effects of band pass filtering, reverberation and noise must not be strongly frequency dependent if RASTI measurements are to be reliable. 4

9 In 2001 a new measure was introduced: The Speech Transmission Index for PA systems (STIPA). Its novelty lies in its use of all seven octave bands and all fourteen modulation frequencies. By using two different modulation frequencies in each octave band the number of data points is kept at fourteen. Speech Intelligibility Index The Speech Intelligibility Index (SII) is similar to AI as it measures the signal-to-noise ratio over the transmission path. Measurements are made either in 1/1 octave, 1/3 octave bands or other selected bands, depending on the type of analysis. SII has incorporated some vital aspects of STI analysis, most importantly the weighting of certain frequency bands depending on their importance to intelligibility. The importance is given by the information content of each frequency band which may vary with the language spoken or the gender of the speaker. Different weighting factors exist, however an average is also provided. Improved procedures compared to AI measurements make measurements in reverberant fields possible [6]. These methods involve an MTF approach as seen in the STI description. For a detailed description of the SII measurements performed in this work, see section 3.4. below: A summary of the most commonly used speech intelligibility parameters is given by Table 1 Table 1: Speech intelligibility parameters [7, 3] Speech Transmission Index STI Weighted signal-to-noise ratio Room Acoustic Speech Transmission Index RASTI Simpler version of STI Speech Transmission Index for PA systems STIPA Simpler version of STI Speech Intelligibility Index SII Weighted signal-to-noise ratio Privacy Index PI P I = (1 SII) 100% Larm and Hongisto [6] have performed an experimental comparison between STI, RASTI and SII. In their study they used both the MTF approach to SII and the pure application of measured speech and noise levels. The latter is of special interest for the study undertaken here, since this was the SII version applied. Larm and Hongisto comment on that the ANSI S standard states that the determination of SII values with this simple procedure is allowed in low reverberation conditions without providing reference values for low reverberation. They performed measurements in seven different reverberation conditions and found that only for measurements in a hemi anechoic room the simple procedure correlated sufficiently with the MTF approach. However, this is only crucial for very high signal to noise ratios, since the SII approach without the use of MTF treats a high signal to noise ratio automatically as a measure 5

10 of good intelligibility. MTF based methods can on the other hand detect damaged modulation, which would cause low intelligibility also when the signal to noise ratio is high. However, when the signal to noise ratio is low, the intelligibility will be low as well no matter what the modulation depth of the signal looks like. For reference, see Figure 12 in [6]. Also, the room condition closest to the hemi anechoic room but with more reverberation tested by Larm and Hongisto is already considerably more reverberant than the room tested in this work, see section Simulation Geometrical room acoustics Geometrical room acoustics is an energy-based method of describing acoustic systems. Its basic principle is the description of the sound field not taking the wave character of sound into consideration but rather viewing it as consisting of independent sound rays. These can locally be regarded as traveling plane waves [8]. In such an energy-based method the acoustic field is characterized by its total energy content and the sound rays by their energy density or intensity. If several sound sources act on the system their influence in some point can be treated as a superposition of sound power, since the sources are regarded as uncorrelated. It should be noted that geometrical room acoustics treats the initial development of a sound source being switched on. Furthermore can phenomena like interference and diffraction not be treated by geometrical room acoustics, since they depend on the wave-character of sound. Rindel [9] gives a good summary of how the ODEON ray tracing software used in this work makes use of a Hybrid Method, that combines two approaches to geometrical room acoustics: The Image Source Method and the Ray Tracing Method. Image Source Method surface. This method is based on mirroring the source in plane of the reflected A great advantage of this method is that it is very accurate but only realistically applicable in simple room shapes. Reasons for this are the great number of image sources that need to be determined if the room is modeled by many surfaces, given by N sou = 1 + n [ ] (n 1) i 1 (n 1) i, (3) n 2 where i is the reflection order and n the number of surfaces. However, many of these image sources will not be relevant for a specific receiver position of interest, as can be seen when looking at approximate number of image sources within a radius of ct N ref = 4πc3 3V t3, (4) 6

11 where c is the speed of sound in air, t is time and V the volume of the room. Note that this approximate number is obtained for a simple rectangular room, but is statistically valid for any room geometry. As an example, Rindel compares the number of possible image sources N sou to the number of image sources valid for a specific receiver N ref for a room with volume 1500m 3 modeled by 30 surfaces after 600 ms: N sou of which less than 2500 are valid for the receiver, stating that most of the calculation effort of finding the image sources will have been in vain. Ray Tracing Method This method treats sound from a source as a number of particles emitted from the source point. As the particles travel through the room, they loose energy when reflected from surfaces according to the surface s absorption coefficient. The directions of propagation after a particle has been reflected are determined by specular reflection. To be able to determine whether the particles find the position of the receiver, the receiver position is either defined as a volume or an area around the receiver point or the sound rays are modeled as the axis of pyramids. The Hybrid Method The idea that creates this method is that when combining the two aforementioned methods a so called visibility test is performed. This is doen by backtracing the image sources from the receiver position. If the image source is found valid, i.e. visible, the level of the reflection is given by the product of the absorption coefficients of the walls hit and the level of the source. The time of arrival is given by the distance traveled. To avoid duplicates, i.e. rays that discover the same image sources, an image tree needs to be stored ODEON Ray tracing software The ODEON software allows a variety of source receiver combinations, e.g. point, line or surface sources. In the following the used feature of point response calculations is described briefly in order to specify how ODEON makes use of the theory shown above. This part is based on the ODEON manual [10]. The point response calculation in ODEON is divided into two parts: A receiver independent and a receiver dependent part respectively. Note that this is done only to save calculation time as some calculations can be reused. The distinction between image sources and secondary sources in the ODEON software is due to ODEON s secondary source method which makes it possible to generate a reverberation tail [9]. Receiver independent calculations Early part: Up until a specified transition order (TO) ray reflections are treated as specular reflections. The program then creates image sources. Rindel [9] states that experiments show 7

12 that an optimum TO is found to be two or three. Late part: Reflections are treated according to vector-based scattering and secondary sources are created if the number of reflections the rays have performed is above TO. Receiver dependent calculations Early reflection method: The program checks whether the image sources are visible from the receiver position. If so, the reflection is added to the reflectogram. In addition each image source is split into a specular and a scattering contribution. The latter is denoted as early scattering and early scatter rays are traced up to TO. Here the scattering coefficients specified by the user are used and other phenomena such as edge diffraction are taken into account. Late reflection method: The program checks whether the secondary sources are visible from the receiver position. Again, if so is the case the reflections are added to the reflectogram. The secondary sources are given directivity factors according to Lambert, Oblique Lambert (ODEON default) or uniform directivity. The user can specify which of these directivities are to be applied. 2.3 Using screens Screens in room acoustics Wang and Bradley have in their two successive articles given an elaborate description of a mathematical model for the prediction of SII properties for a single screen [11] [12]. They point out that the two mechanisms via which speech reaches a receiver on the other side of the screen are diffraction and reflection over the screen. An important issue pointed out is that their first model for calculating the IL for a single screen in a flat room took interference between sound waves into account, which requires known acoustic impedance values for the ceiling and floor. However, since interference effects usually are of significance only in the lower frequency range they can be disregarded as the sound energy is of lesser importance to speech privacy in that regime [ANSI 3.5 (1997)]. Based on this consideration a simplified model based on energy summation is introduced. The authors provide a study of the effects of a number of office parameters on SII. The parameters found to have the greatest influence on SII are the ceiling absorption and the screen height. Furthermore, it is concluded that these two parameters interact and that good privacy (i.e. low SII, high PI) can only be achieved by both high screens and highly absorbing ceilings Dealing with diffraction In the case of a train compartment there are three mechanisms that contribute to a sound field behind a sound barrier: Transmission through the barrier, reflection from surfaces around the source/receiver and diffraction at the barrier edge. Numerous studies have been published 8

13 about sound barriers for use at motorways. In that case only transmission and diffraction apply. Much effort has been undertaken around understanding the diffraction mechanism and thereby finding possible solutions to minimize its contribution to the sound field in the shadow region [13, 14, 15, 16, 17]. Some contradicting conclusions have been reached on whether absorbing edge configurations actually manipulate the sound progression by diffraction in a beneficial manner or not (Yamamoto [17] vs. Möser [15]). Most conclusive seems to be that an end piece needs to be designed with the parameter of acoustic impedance rather than absorption, as done by Möser [15] and Okubo et al. [14]. A conclusion of Möser s accounts on the effect of cylindrical head pieces for noise barriers [15] is: No matter how the sound propagation around the head piece is described, either as tangential guidance of power or parallel volume flow, both descriptions state that the incidents on the cylinder surface form a sound source that acts on the shadow region. The impedance must be chosen in a way that only small intensities arise at the surface. Two extreme cases are shown: Head pieces with zero and infinite impedance: In the zero impedance case a boundary layer forms a non-compressed and freely moving mass. Locally, the displacements of this mass imposed by the forces of the passing wave are governed by the law of inertia: Displacement and force are out of phase, resulting in high pressure at the outer boundary of the mass as it is displaced outwards, see Figure 2a. Thus the pressure field cannot cling to the cylinder surface and the boundary layer does not transport any power. In the infinite impedance case however the boundary layer cannot dodge the pressure field by an inwards motion and has to react with compression. Thus the wave fronts can cling to the surface as they pass by and also stretch into the shadow side of the cylinder, see Figure 2b. 9

14 (a) (b) Figure 2: (a) Particle movement at the head piece with zero impedance [15]. movement at the head piece with infinite impedance [15]. (b) Particle 10

15 3 Method 3.1 The train compartment model Measurements were performed in a full scale mockup model of a train compartment. Simulations were performed with the ODEON ray tracing software The mockup The mockup was placed in the hemi-anechoic room at MWL and a set of two seats placed in it, see Figure 3. The front and back were kept open to represent infinite lengths. In order to control reflections at the walls and ceiling, absorbing panels were mounted. These panels had very high absorption above 500 Hz, so full control of standing waves in the low-frequency regime cannot be assumed, but the high frequency modes of the untreated were successfully damped. The material used for walls, ceiling and half of the floor was pressboard, the other half of the floor was open and the concrete floor of the hemi anechoic room was visible. The pressboard had 10 mm thickness and was at the roof and ceiling backed by a 45 mm cavity and a 3 mm pressboard layer. The half of the floor that was wooden was a 10 mm pressboard layer backed by a 100 mm cavity and the rigid floor. The dimensions of the whole mockup were (l w h) = ( )m. Figure 3: The mockup model in the hemi-anechoic room at MWL, KTH. Its dimensions are (l w h) = ( )m. 11

16 3.1.2 The ray tracing model An ODEON model was created using SKETCHUP and exported to ODEON using the ODEON plugin. The geometry of rooms used in ray-tracing software needs to be simple, since using too many surfaces will not combine with the image source theory applied by ODEON for early reflections, see section Thus the geometry of the ODEON model is kept simple, see Figure 4. Figure 4: The ODEON model, geometry based on the mockup as seen in Figure 3. The total number of surfaces of this setup (i.e. no screens) is 27. Table 2: ODEON model surfaces and properties Description Area [m 2 ] Absorption coefficients [-] in the 1/1 octave bands with center frequency [Hz] visible wooden walls visible wooden ceiling absorber plates wooden half of floor concrete half of floor back rests seating surface back rest top areas seat sides seat lower fronts seat backsides open front/rear end In order to verify the ODEON model, measurements of reverberation time (T 60 ) were performed. Because the room under consideration was highly absorbent considering the the open 12

17 ends of the mockup, which were modeled as totally absorbing in the ODEON model the values of T 60 were expected to be low. Since no reflection was occurring at the open ends and all sound emitted from a source towards these ends was expected not to be reflected back to a receiver on the same y axis, no complete room average was measured, but rather an average of a plane, i.e. source and receiver on the same y axis, see Figure 5. Measurements in the mockup model were performed using the ARTA software. It excites the room with a logarithmic frequency sweep in order to subsequently measure the decay of sound power using the T 30 method. Results from this measurement are shown in Figure 6a. Figure 5: The ODEON model with receivers (blue dots number 8,9 and 10) and sources (red circles P6 and P7). Note that all sources and receivers are on the same y coordinate. The plane spanning this y coordinate is the only part this model is verified for. Since the geometry was given by the mockup and high detail accuracy of the ODEON model is no way of improving the results of the simulation, the only remaining parameter was the choice of material assigned to the surfaces. Such a choice may set a value for the absorption and scattering coefficients. Only for the absorber plates the absorption coefficient was given by the producer. All other materials needed to be selected from the ODEON material library or set manually. The procedure towards a good match between measurements and simulation was as follows: Materials whose descriptions seemed to match the actual physical setup (e.g. plywood paneling with hollow cavity) were chosen and the simulation run. If the match in T 60 was OK but improvable, the material properties were changed accordingly, setting out from the values used until then. As an example, the frequency dependency of the absorption factor could be altered. The final material properties used for simulations are shown in Table 2. Results from the final ODEON setup are shown in Figure 6b, and a comparison of the final ODEON setup and the measurements is shown in Figure 6c. 3.2 The artificial human head An artificial head was built to be able to introduce the speech signal with a directivity pattern more closely related to the directivity of human speech than a regular monopole speaker would 13

18 ReverberationMTimeMT 60 [s] FrequencyM[Hz] S1MR1 S1MR2 S1MR3 S2MR1 S2MR2 S2MR3 Mean ReverberationMTimeMT 60 [s] FrequencyM[Hz] S1MR1 S1MR2 S1MR3 S2MR1 S2MR2 S2MR3 Mean (a) Measured reverberation times (ARTA software) (b) Simulated reverberation times (ODEON software) ReverberationOTimeOT 60 [s] FrequencyO[Hz] ARTA ODEON (c) Comparison of mean reverberation times Figure 6: Results from the ARTA reverberation time measurements and ODEON simulation (S = Speaker, R = Receiver). allow. It is shown in Firgure 9 below. Directivity measurements were conducted to verify its pattern. The frequency range was 250 to 4000 Hz and the angles 0 to 180 degrees with 30 degree steps in the horizontal plane. The signal was pink noise and well above the background noise in the hemi anechoic chamber. The distance from the artificial head s mouth to the microphone was 1 m. Note that during these measurements the microphone was kept in a fix position and the artificial head turned stepwise around its vertical axis. This caused the distance between the artificial head s mouth and the microphone to increase with up to 180 mm, which is the length of the artificial head. The measured SPL were inserted into ODEON s plot editor tool to create a directivity pattern that equals the directivity of the artificial head, see Figure 7. The octave bands 125 and 8000 Hz were given the same directivities as the 250 and 4000 Hz octave bands, respectively. The directivity pattern created was applied to a point source in a free field environment in ODEON and the SPL simulated at 1 m in front of the source. These SPL were compared to the measured SPL as the artificial head was fed with filtered pink noise (where the signal was aimed at resembling the standard speech spectrum level given in the standard). The difference between the simulated and the measured SPL was applied as an equalizer on the ODEON directivity pattern. Thus this pattern should have the same shape as the measured directivity pattern and produce the same SPL at 1 m in front of the source as the artificial head. This 14

19 was confirmed successfully in another ODEON free field simulation. For reference a directivity pattern available in the ODEON directivity files library is shown in Figure 8. Note that the ODEON directivity pattern is defined in 3D, whereas the pattern created in ODEON from the 2D directivity measurements in the horizontal plane on the artificial head may be defined in 3D as well, but is uniform in the vertical plane. Thus, the simulated artificial head directivity pattern resembles the actual head correctly in the horizontal plane, but not necessarily resembling it correctly vertically. (a) 250 Hz (b) 500 Hz (c) 1000 Hz (d) 2000 Hz (e) 4000 Hz Figure 7: Directivity pattern of the artificial head. Only the horizontal plane is shown, with the front (0 degrees) pointing upwards in the figure. Shown are the measured octave band patterns after they were inserted into ODEON and interpolated. Note that the levels shown are not the measured levels, but the levels after an equalizer was applied to obtain the speech spectrum levels according to the standard. 3.3 The screen prototypes Two screen prototypes were constructed out of 6 mm thick acrylic glass equipped with absorbing foam with 30 mm thickness, see Figure 9. The prototype that in the following will be denoted as the full box had four screens: On the two sides, the top and the rear. The reduced box prototype had three screens: On the left side of the screened passenger, the top and the rear. In the ODEON model the screens were given the absorption coefficient of window glass and sound 15

20 (a) 250 Hz (b) 500 Hz (c) 1000 Hz (d) 2000 Hz (e) 4000 Hz Figure 8: Directivity pattern of normal vocal effort speech in ODEON. The same plane as in Figure 7 is shown. reduction index through the screen according to the mass law. Measurement and simulation comparison showed that the reduction index was not correctly estimated, which most certainly is because the screens of finite size and the influence of free bending waves to the sound radiation of the screen or the boundary conditions due to to mounting of the screens. The screen insertion loss (IL) was both measured and simulated for the full box case and the screen reduction index according to the mass law calculations altered to acrylic glass with 3 mm thickness. Results of those measurements and simulations are shown in Figure 10. It seems that a change in reduction index which is the effect of a change in thickness has little effect on the behavior of the shielding performance. The screen s material properties as they were inserted into ODEON are shown in Tables 3 and 4. 16

21 (a) The prototype full box with four (b) The prototype full box. Setup case: screens (left, right, back, top). Setup case: Speaker in the box, receiver on the neighreceiver in the box, speaker on the neigh- boring seat, i.e. outside. boring seat, i.e. outside. (c) The prototype reduced box with (d) The prototype reduced box. Setup three screens (left, back top). Setup case: case: Speaker in the box, receiver on the Receiver in the box, speaker on the neigh- neighboring seat, i.e. outside. boring seat, i.e. outside. Figure 9: The two prototypes used, mounted on a seat. Also shown are the four possible measurement setups with the speaker (artificial human head) and the receiver (microphone) positions. 17

22 IL [db] ODEON IL 6mm screens ODEON IL 3mm screens MEAS. 6mm screens Frequency [Hz] Figure 10: The simulated insertion loss (IL) for two acrylic glass thicknesses: 6 mm and 3 mm. Also the measured 6 mm case is shown. Table 3: Screen properties in the ODEON model Description Area [m 2 ] Absorption coefficients [-] in the 1/1 octave bands with center frequency [Hz] side screens top screen rear screen Table 4: Screen sound reduction in ODEON Center frequency [Hz] Sound reduction [db] Center frequency [Hz] Sound reduction [db]

23 3.4 Measuring SII Measures of speech intelligibility were described in section 2.1. Here the SII according to the standard ANSI S is used. The SII calculation according to the ANSI standard is in this work done with the speech spectrum and train cabin noise spectrum (both measured in 18 1/3 octave bands from 160 to 8000 Hz at the center of the listener s head position) as indata. In this section a short summary of the steps leading from this indata to the final single SII value is given, followed by a table serving as an illustrative example with all the calculated values for each step for a real measurement performed in this project, see Table 6. The indata is in the standard denoted as spectrum level of equivalent speech (E ) and spectrum level of equivalent noise (N ), respectively. Note that the standard could take hearing aid or hearing protection devices into account. If neither are used, as was the case in this study, the values of E and N are equal to the measured values. Further data needed for the calculation are the third octave band center frequencies (F) with numbers i and the equivalent hearing threshold level (T) which was set to 0 db in all 1/3 octave bands. The remaining 13 values needed to obtain the SII index are denoted SII calculation parameters in Table 6. They are either calculated from the measured indata or tabulated values defined in the standard and described in Table 5. Note that the band importance functions (I) can be changed to fit the situation of the measurement. The default functions are for average speech and were used here, but other speech tests are given, e.g. various nonsense syllable tests (NNS), Diagnostic Rhyme Test (DRT) and more. For further descriptions of the SII calculation values, see the ANSI S standard. 19

24 Table 5: SII calculation parameter summary V self-speech masking spectrum level V i = E i 24 [db] the larger of the equivalent noise B spectrum level, N i or the self speech C masking spectrum level V i max(n i, V i) [db] the slope per octave of the spread of masking due to band i Z equivalent masking spectrum level X X D U L K reference internal noise spectrum level equivalent internal noise spectrum level Spectrum level for equivalent disturbance standard speech spectrum level at normal vocal effort level distortion factor temporary variable C i = 80 + (B i + 10 lg F i 6.353) Band 1: Z 1 = N 1 All other bands: [ ( ( ))] Z i = 10 lg N i + i 1 k B k +3.32C k lg 0.89 F i F k given in the standard, see Table 6 X i = X i T i max(z i, X i ) given in the standard, see Table 6 L i = 1 (E i U i 10) /160 L i = min(1, L i ), i.e. limited to a value from 0 to 1 K i = (E i D i + 15) /30 L i = min(1, L i ), i.e. limited to a value from 0 to 1 [db/octave] A band audibility function A i = L i K i [-] I band importance function given in the standard, see Table 6 [-] SII speech intelligibility index SII = n i=1 I ia i [-] [db] [db] [db] [db] [db] [-] [-] Table 6: SII calculation example according to ANSI S i F i E i N i T i V i B i C i Z i X i X i D i U i L i K i A i I i I i A i Indata SII calculation parameters SII =

25 4 Measurements and simulations 4.1 Measurements on the acoustically shielding box Measurements were performed on the two screen prototypes described earlier. The aim was to determine this design s performance in screening against train cabin noise and speech to subsequently draw conclusions about the impact the screens have on speech intelligibility. The Measurements were carried out and evaluated in accordance with the ANSI S standard, which gives values of speech intelligibility as the SII, described in section 3.4. Measurement setup The measurements were carried out in the mockup model shown in section The train noise was introduced through a 5.1 surround sound system according to Figure 11. Figure 11: Positions of the five speakers and the subwoofer (shown as a larger speaker) in the mockup model. The distance to the floor of all speakers was 1.2 m, the subwoofer stood directly in the floor. This setup was used exclusively to introduce the train noise signal. The speech signal was introduced through the artificial head speaker described in section 3.2. According to the speaker/listener position cases, the speaker could be placed on either of the two seats. The measurements were carried out with a BSWA 1/2 in. microphone connected to a PC through an external sound card and recorded with the SPECTRA PLUS software. Train cabin noise and speech signals The train noise signal was a real life recording from a moving train in regular traffic and was played back as described above. The total SPL at the listener s position in the mockup (without any obstructions by screens) was matched with 21

26 SPL [db] measurements obtained from a train traveling at 180 km/h. The spectrum measured in the mockup also showed acceptable congruence with the measurements from the traveling train, see Figure Measurement at 180 km/h Mockup measurement Measurement noise Frequency [Hz] Figure 12: The train cabin noise as it was measured on the train in regular traffic and how it was resembled in the mockup. Also the background measurement noise is shown, see the Remarks paragraph. The speech signal spectrum was given in the standard, which provides spectra for four vocal efforts: Normal, Raised, Loud and Shout. The normal vocal effort spectrum was used. The signal was generated by a noise source that produced pink noise and sent through an equalizer and an amplifier before being sent to the artificial head speaker. The equalizer and amplifier were adjusted to produce a spectrum which for verification was measured in free field conditions at one meter distance from the speaker, as stated in the standard. Once the noise source, equalizer and amplifier were correctly adjusted, they remained unchanged and the artificial head speaker could be mounted in the mockup. Remarks The equalizer was not able to attenuate the pink noise signal sufficiently to resemble the standard speech spectrum correctly in some of the higher 1/3-octave bands. Also, there was a substantial amount of thermal noise in the measurement chain, presumably from the sound card. See Figure 13 for both phenomenons. 22

27 SPL [db] Standard spectrum "normal" 1 m free field Measurement noise Frequency [Hz] Figure 13: The standard speech spectrum for normal vocal effort which was to be produced by the artificial head at one meter distance in free field conditions. Also the background measurement noise is shown Results The results shown in this section are given for the frequency span that SII is defined for, i.e Hz. Further are all SPL given in A weighted decibels, i.e. db(a). This is to make the impact of the SPL on acoustic comfort more obvious. The measurements of the train cabin noise show that the two prototypes have some effect in shielding against the noise introduced to the cabin from numerous directions, see Figure 14. In all but the two lowest 1/3 octave bands and the 800 Hz band both screens attenuate the noise. The total SPL in db(a) shows that the full box gives a reduction of 4 db(a), whereas the reduced box gives 0 db(a), i.e. no reduction. This is due to the lack of reduction in the lowest frequency bands, as the high SPL in these bands dominate the total SPL. Note that in the highest bands the signal to noise ratio considering the measurement noise (see Figure 13) is critical. The speech signal measurements, see Figure 15, show that the screens perform in a similar way as in the train cabin noise measurements, see Figure 14: The lowest 1/3 octave bands are not attenuated, as are some bands around 500 to 800 Hz. Regarding the two cases of the 23

28 70 60 BackgroundgtraingnoisegwithgandgwithoutgthegfullRreducedgbox Cabingnoiseg69dBCg60dBlA/ Fullgboxg67dBCg56dBlA/ Reducedgboxg71dBCg60dBlA/ 50 SPLg[dBlA/] Frequencyg[Hz] Figure 14: The train cabin noise measured in 1/3 octave bands ( Hz) at the center of the receiver s head position. The three cases are: (solid line) cabin noise without any screens, (thick dashed line) cabin noise inside the full box and (thin dashed line) cabin noise inside the reduced box. The two values given in the legend are the total SPL (linear) and the total A weighted SPL in the shown frequency range. speaker/receiver positions (inside or outside the box), some degree of reciprocity is obvious. Keeping track of the SII weighting factors allows for quick estimates of the importance of attenuation in certain frequency regions, e.g. the poor performance of both prototypes in the lowest frequency bands will have little influence on speech intelligibility, as this region is not strongly weighted. Figure 16 shows no results that have not been defined previously. However, it shows the critical step in the evaluation of the prototype s performance in altering speech intelligibility, as it shows the two signals who s superposition the receiver will experience in each of the four possible setups. These setups are: 1) The receiver being in the full box and the speaker on the neighboring seat, 2) the speaker in the full box and the receiver on the neighboring seat, 3) the receiver in the reduced box and the speaker on the neighboring seat and 4) the speaker in the reduced box and the receiver on the neighboring seat and are all shown in Figure 9. In each case a unique combination of the results shown previously in this section is compared and an evaluation according to the SII calculation performed. 24

29 Receiversinsboxs67dB,s59dBvAb,sSII=pI7 Speakersinsboxs68dB,s6kdBvAb,sSII=pI57 Nosboxs7pdB,s66dBvAb,sSII=pI7l WeightingsfactorssinsSIIscalculation 7q Receiver=in=box=7qdBl=6HdBhAtl=SII=q86z Speaker=in=box=7qdBl=6zdBhAtl=SII=q857 No=box=7qdBl=66dBhAtl=SII=q875 Weighting=factors=in=SII=calculation 6p 6q 5q SPLs[dBvAb] lp SPL=[dBhAt] 4q zq 8p pik pip5 Weightingsfactors Hq q8y q8q8 q8q6 q8q4 q8qh Weighting=factors kp 8 kp N kp lp Frequencys[Hz] (a) yq H yq z yq 4q Frequency=[Hz] (b) Figure 15: The speech signal measured in 1/3 octave bands ( Hz) at the center of the receiver s head position. Results from the two prototypes are shown: (a) full box (b) reduced box. The two possible speaker/receiver position cases are: (dashed line) receiver in the box and (thin solid line) speaker in the box. Also the case of no implemented screen is shown (thick solid line) and the weighting factors of the SII calculation (dash-dotted line), see band importance function in Tables 5 and 6. Note that the weighting factors are plotted against the right Y axis. The total SPL values are given as in Figure 14 and the SII value according to the calculation described in section ODEON simulations ODEON ray tracing simulations were performed in the model described in section The version of ODEON used in this work (ODEON Industrial) returns results in 1/1 octave bands only. In the following figures ODEON simulated results and measured results are compared. The measured results are the same as shown in 1/3 octave bands section but here shown in 1/1 octave bands. Note that therefore the frequency range shown in the figures in this section is Hz and not Hz, as in the previous sections. This is due to the 160 Hz 1/3 octave band belonging to the 125 Hz 1/1 octave band. Note that the calculated SII values shown in this section are as before based on 1/3 octave band calculations as described in section 3.4. The source and receiver positions were according to the measurement setups shown in Figure 9. The speech signals were introduced by a point source with directivity and sound power based on the measurements performed on the artificial human head, see section 3.2. The prototype s properties were inserted into ODEON according to Tables 3 and 4 in section 3.3. The 25

30 Noise SpeechISII=Wa7 SIIIweights ReceiverIinIbox Noise SpeechISII=Wa57 SIIIweights SpeakerIinIbox 6W 6W SPLI[dBFAq] 4W 2W Waf WaW5 fw 2 fw 3 fw 4W FrequencyI[Hz] Noise SpeechhSII=8a63 SIIhweights (a) 4W 2W Waf fw 2 fw 3 fw 4W FrequencyI[Hz] Noise SpeechhSII=8a57 SIIhweights WaW5 WeightingIfactors Receiverhinhreducedhbox 78 Speakerhinhreducedhbox SPLh[dBFAq] 48 f8 8aW W8 f W8 3 W8 48 Frequencyh[Hz] 8a85 (b) f8 W8 f W8 3 W8 48 Frequencyh[Hz] 8aW 8a88 8a86 8a84 8a8f Weightinghfactors Figure 16: The speech signal (thin solid line) and train cabin noise (solid line) measured in 1/3 octave bands ( Hz) at the center of the receiver s head position and (against the right Y axis) the weighting factors of the SII calculation (dash-dotted line). (a) full box, (b) reduced box, with the two possible speaker/receiver position cases, respectively (as seen in Figure 9). train cabin noise was not simulated, since the ODEON model only had been verified for the plane going through the row of seats. Therefore the measured train cabin noise values for the differ- 26