When considering how to model a room in a computer, several questions come to mind regarding the accuracy of both the modeling software, and the model itself. Room analysis software is nothing new, but given the fact that computers are becoming faster and able to store more information, this software is now being used everyday for analysis, refinment, and design of spaces requiring good acoustics. Given that most programs that do this type of work rely on a variety of methods, including hybrid calculations, one is bound to ask which algorithms are the best. Michael Vorlander addressed these concerns in a 1995 paper titled "International Round Robin on Room Acoustical Computer Simulations", determining that among several commercial analysis programs only three were reliable in predicting room behavior. Assuming one possess an accurate program, concerns regarding model complexity become an issue. This paper addresses this issue through the use of increasingly complex computer models of the Penn State Music Recital Hall and a correlation analysis of impulse responses from the RAYNOISE room modeling program and experimental data.
The intended objectives were as follows:
Although not of high complexity (lacking any but very basic and dominant room structures), the models required for this project consumed an amount of time sufficient to slight the last objective, and interfere with the third. Additionally, RAYNOISE has problems writing database files in some cases, and this made computations difficult.
Graphics of the modeling blocks of the room can be found at the URL:
http://www.acs.psu.edu/users/smithsh/hall_acoustics.html
Seating 500 from plans - counted at 480 Front row of seats has been removed. Length 89.5 feet Height 32.0 feet from stage platform/back wall Width 60 feet Stage depth 30 feet Audience seating area depth 52.5 feet Angle of Stage walls 10 degrees Angle of overhead reflector 8 degrees above horizontal Floor: Exposed concrete and linoleum over concrete. Utility carpet in aisles Walls: painted concrete block with adjustable or plaster over concrete Back Wall/Wedges: plaster under acoustical cloth screen Ceiling: Painted precast concrete tees Stage Walls: Varnished wood Stage Reflector: Plaster on metal lath Seats: Medium thickness upholstery with metal pans and sides (including wooden writing tablets). Usage: Music department student and faculty recitals. Medium size performing groups such as choir and chamber and jazz ensembles, small orchestras. Note: The dimension ratio: 1:2:2.8 does not conform to any of the favorable room dimension ratios (Sepmeyer, Louden, Volkmann, Boner - see Everest p231).
Each is a major feature of the room, and should contribute to the acoustics. In theory, the response of a modeled room should approach that of the actual room given 1) that the modeling software is reliable, and 2) that the features being added to the room contribute to the room's behavior. The analysis was conducted without including the velvet curtains, as if they were retracted, and also assumed that the grille cloth is acoustically transparent. This is because although RAYNOISE can simulate sound transmitted through one barrier to another, it cannot compute sound transmitted back through the same barrier (i.e. sound transmission is one way). In addition, the dimensions of the pipe organ were unavailable, so only a rough approximation is used, which is actually about 75% of the size of the actual organ. The measured data from the hall used in this study was taken with the curtains fully drawn back.
Room models were constructed using I-DEAS modeling software and methods previously employed in ACS597D. Due to the complexity of I-DEAS it is quite likely that there are other methods and programs that those used here which can be employed, but it recommended the persons using this program without prior experience stick to established methods for modeling objects, applying what they know.
Experience plays a major role in designing any model with I-DEAS, and this case was no exception. One should choose their construction method based on how they intend to join various parts of the model, and whether or not they intend to mesh it automatically or manually. Automatic meshing of a room is a quick and dirty way in which to do very rough modeling, but it in no way creates surface elements that would resemble real-world structures. For example:
1)Automatic meshing generates grids containing elements of strange shapes and sizes. In order to model a room, material characteristics must be applied to these elements. Real world structures rarely if ever contain shapes other than rectangles, cylinders and spheres. The irregular meshing makes it difficult to define the location of a regularly shaped object such as a window, or a wall.
2) Automatic meshing can result in too many elements per surface. Real world walls consist of one element, so why model them with 100 smaller elements in a mesh? Again, material properties must be applied to each element, and in RAYNOISE this is done either manually, or by using the unreliable RAYNOISE GUI. An automatically generated mapped mesh can be used only in the case that regular variation of materials occurs across a surface, such as art-deco brick and glass, and most likely one would model this type of wall as one element with unique absorption coefficients.
There are other methods for obtaining a regular mesh of a reasonable number of elements across a surface, such as importing a volume mesh into a program like SYSNOISE and saving the exterior elements as a second mesh. But the size of the volume elements need to be as small as the smallest feature on the model, or that feature will not be meshed properly. Again, one ends up with too many elements on a surface.
The best way mesh up a room is manually. However, one needs to build up a model of the room before a mesh can be created. There are certain considerations that must be followed in modeling the room with I-DEAS.
I-DEAS != IDEAL
I-DEAS is a program designed for the modeling, construction, visualization, and analysis of mechanical systems, and incorporates several analysis tools, including meshing algorithms. However, this program has been designed to be used in mechanical engineering, in which parts interact, translate, and vibrate. It can be used to design rooms, but this was not its intent. The model of a room is best constructed by a dedicated CAD program which does not incorporate the same level of complexity and creates the mesh during the design process. Because it was designed to study the effects of vibration on mechanical systems consisting simple, single piece constructions (via Finite Element models), I-DEAS requires that all parts of the model be joined into one part before the model can me meshed either automatically or manually (here joining refers to the physical incorporation of two parts into one). What this means is that in order to get a model of a room from I-DEAS to RAYNOISE for analysis, one must design individual features either from the same piece of material (one part), or join several pieces of material together before meshing.
In addition to being able to only mesh one part at a time, some additional constraints are placed on design methods along the way:
Although one can design models out of different parts, resulting in separate part meshes, this won't work for two reasons:
1) If the parts are joined in I-DEAS, automatic meshing will result in too many small elements being generated.
2) Individual part meshes cannot be joined in RAYNOISE. RAYNOISE will work with only one mesh at a time, so joining any separate meshes (walls, stage, etc.) into a single room model in RAYNOISE won't work.
After several failed attempts at solutions to the above problems, one method stands out above all so far for construction and meshing of a room with I-DEAS: Design all room features to be cut from a solid block which will become your meshing part. Cut them from the block, and mesh the block. A CAD program would do exactly the opposite: building a room from many different plane pieces, simultaneously making a mesh of them, which is similar to building a house of cards. This approach has not been tested in I-DEAS, but the program should be able to accommodate it.
The best way to think about building a room with I-DEAS is by considering it to be a block of metal. I-DEAS works with shapes. It saves a number of basic shapes and operations on those shapes to build models. This means that the data structure of a complex part in I-DEAS is really the location and dimension of several basic shapes, not nodes and planes (like a CAD program would store information). By "machining" parts of metal out of your block, you create a "negative" of the object you want.
In Summary:
I-DEAS: PART ---\
\
PART -------> MESHING PART ---> MANUALLY MESH ---> EXPORT
/
PART ---/
CAD: PARTS ------> SAVED AS FILE WITH AUTOMATICALLY ASSIGNED ELEMENTS & NODES
This method was only arrived at after several weeks of working with I-DEAS, and this is where the experience factor comes into play. No less than three different "phases" of model construction were completed. The first phase consisted of individual parts to be joined into one model after meshing separately. The second phase of model construction was done to cut the features from one block and have I-DEAS automatically mesh up the resulting model. Due to small features that would have resulted in both unusually shaped elements and elements that were too small, this method was abandoned for phase three. Phase three consisted of models that used cutting parts to make even simpler models from a single block. An additional reason for this simplification is because it became necessary to manually mesh the models (for reasons previously described, and continued below).
To further complicate meshing up a room in I-DEAS, existing faces often have to be divided up into smaller quadrilateral and triangular elements. An example is having to divide up the face of an L-shaped object into 2 separate elements. Since a model may contain many of these pieces the number of elements may be 2-3 times the number overall faces in the model. Again, computer meshing is undesirable because of the large number of elements that will need to be manually assigned materials in the room simulation program, and because of the element shape/orientation. Then how does one ever select elements in RAYNOISE? The element selection is impossible! So, one doesn't want to manually complicate the mesh in a manner similar to the way the computer would if one doesn't have to.
Here are some additional considerations when building a room model for meshing in I-DEAS:
RAYNOISE combines acoustical ray tracing and mirror-image-source methods for room analysis. This process requires defining sources and receivers of sound in the enclosed space. For low frequencies, mirror-image sources are created in each plane of the room, and are treated as band-limited sources who's sound fields at the receiver superimpose. Before such a source is used in a calculation, a visibility test is performed to determine if the mirror image has an unobstructed transmission path to the receiver. If not, that source is not used. High frequency radiation is cast in a uniform distribution over a specified emission angles from the source, for which the user may specify a directivity function. These rays are propagated until one of the following occurs: 1) The attenuation from bounce-absorption causes the level of the ray to fall below a specified limit. 2) A maximum number of bounces from reflective surfaces occurs, or 3) The propagation time limit is reached. The sound at the receiver is determined by how many rays cross the receiver volume.
The two methods RAYNOISE uses for the propagation of rays are the Conical and Triangular Beam Methods. The conical method emits a large number of cones from the source. The frontal area of the cone obeys spherical spreading principals. Once a receiver is found to be within the volume of a cone, the pressure at the receiver is derived from the spherical spreading. In order to insure a spherical wavefront, weighting functions are applied to the font of the cone, and cones are propagated such portions of them overlap. The triangular beam method uses pyramids to discretize the wavefront, with the apexes of these pyramids being located at the source. This method requires no overlapping, and is considered more accurate.
RAYNOISE also allows for statistical principles of energy emitted from false sources to be used to account for any left over energy from discontinued rays (described above). The program also has an option to use random diffusion at reflection surfaces. For each bounce, a random number is generated and if the number is lower than the diffusion coefficient of the material a secondary source created at the bounce location gives up some of its energy to all the receivers and generates a new ray propagating in a random direction. Otherwise the reflection is specular. User-defined diffraction edges may also be included in the calculations.
RAYNOISE command files were used for program execution. As will be described later, several different versions of these files were created over a series of data runs with the program. The intention for any computational run is to keep the program parameters constant over all eight models to insure no one model has a computational bias over the others.
Material 4
Name 'cblock painted' Material definition
Sabine 0.11 0.11 0.08 0.07 0.06 0.05 0.05 0.05 Octave band absorption
coefficients from
63 Hz to 8 kHz bands.
Return
Assign
Material 4
Elements 119 125 126 129 131 134 136 187 222 223 Element numbers to
which material 4 is
assigned
Return
It should be noted that segmentation faults will occur if the material
descriptions exceed 41 characters, leading to a fatal error. Fortunately this
only occurs on program exit, and will not stop calculations already in progress.
Source 1
Name 'JBL Eon' Coherent Coherent source
Power 75 80 86 86 84.5 87.5 86 90.5 Power per frequency band
Position 9.371 -4.066 4.5732 Location
Vector -1 0 0 Orientation (to back of hall)
Rotate 0 No rotation
Emission -90 +90 -90 +90 Emission over a half sphere
Alignment 0.0 Zero time delay.
Return
Eventually, due to problems accessing the model databases of the more complex models, the analysis bandwidth of each model had to be reduced to 500 through 4000 Hz in linear steps of 500 Hz. The parameters were initially based on a set of example values given in the RAYNOISE "Getting Started" manual, and are as follows:
Initial Parameters
Rays 5000 Number of rays emitted from the source
ReflectionOrder 40 Max number of reflections allowed per ray
TimeWindow 2000 Time duration for ray propagation
DynamicRange 70 dB SPL at which ray is discontinued
EchoSave 10 Maximum order of echos stored
PathStore 5 Maximum order of reflection for any echo for
which ray path information is stored
StoreLevel 2 Maximum data storage (required for accurate STI
calculation)
HInterval 10 Echo histogram bin interval
HLength 40 Echo histogram bin length
Diffuse 1 Random diffusion algorithm on
Diffract 0 No diffraction edges
Tail 2 Statistical reverberant tail correction
RayMethod 2 Triangular raytracing method
Return
I chose to error on the side of discretion and use both diffuse reflections and statistical tail correction based on the following statement from Vorlander's paper concerning accurate room acoustics simulations:
"It also seems necessary to include diffuse reflections in the algorithms, at least from the 2nd to the 4th order, since the results from purely "specular approaches" were more outlying that algorithms with some kind of random reflections or statistical reverberation tails."
Before proceeding to the analysis, it is necessary to define receivers. RAYNOISE permits one to do this by defining singular fieldpoints, or points distributed over geometrical constructions such as lines, cylinders, etc. I chose to make an analysis mesh which covers the entire audience area, and is defined by four corners at the extrema of the physical dimensions for the ENTIRE audience area (the front corners of the outside sections, and the back corners of the center section). This mesh is a 20x20 grid of elements with an approximate resolution of 0.6 meters (2ft.), and is raised to what can be expected to be approximately the audience ear-level: three feet over the modeled audience area. Points on the grid were chosen to coincide as best possible with the microphone locations during data collection. These locations were reported on a grid having an estimated resolution of 2 feet per element, and therefore the measurements are not extremely accurate as it is. Additionally, the locations of the receivers in the field need not be exact if the field is in fact diffuse.
A sample command file is included in the appendices. Excepting the surfaces to which materials are assigned, the command files used to generate all acoustical data for analysis follow the same analysis procedure, use the same source, and the same analysis parameters.
Impulse response generation requires that one run a mapping analysis and save a database with echograms in it. For optimal results, a store level of 2 with a coherent source is required for any result needing phase information, such as the Modulated Transfer functions used to calculate STI. One would think that using a coherent source with this save option would lead to realistic impulse response generation including phase information, and initially model 4 seemed to yield results of this type. Because the amount of data yielded filled the drive by the conclusion of model 4 analysis, options for all the models were reset (in order to be uniform), and the program was run again. RAYNOISE impulse responses are akin to echograms, in that they contain impulses, constructed of delayed, positive sharp, sinc functions. Initially this was though to be attributed to the fact that the store level was set to 1. However, later calculations with storelevel = 2 and a coherent source yielded the same results. Additional experiments on models provided for the RAYNOISE examples yielded similar results. This leads me to conclude that all binaural impulses generated by RAYNOISE take the form of sinc functions delayed to create an echogram. For this reason, echograms were generated from the experimental data for use in the correlation analysis.
Several problems arose in the process of evaluating these models, especially numbers 5-8. Initially, writing to the database was interfered with by material descriptions that were more than 41 characters in length, but later problems on the more complicated models included problems accessing certain database files for read and write, and floating point exceptions. Since there is no access to the source code, one can only assume that the program may have been addressing unavailable space in a manner similar to that when one reads/writes an array out of bounds in a C program.
The initial intent of the verification analysis for this project was to calculate the correlation of RAYNOISE impulse response data with experimental data collected at six positions in the hall by the ACS597E class using the TEF. These six positions were translated into model coordinates, and points on the RAYNOISE model analysis grid were chosen for impulse response calculations. However, due to constraints on the grid resolution (0.6 x 0.6 meters), correlation analysis was only executed the point that was closest to its real-world counterpart. This point is located 6 feet from the rear wall on the centerline of the hall.
Prior to correlation analysis, code was developed in MATLAB and implemented in a C program to interpolate the physical data from the 48 kHz sampling rate of the TEF unit used for the measurements, to the 44.1 kHz sampling rate of RAYNOISE impulses. A comparison of the interpolated data with the initial data can be seen in the appendicies. The comparison is quite good, with original and resampled data agreeing well over sections observed at high resolution. However, it was deemed necessary that for analysis purposes the data should be filtered to remove any noise artifacts generated by the resampling process.
Of particular note was the execution time of this code in both formats. The same operation which took over three hours in MATLAB with binary data sets was executed with the C program in approximately 2-3 minutes.
Two types of data processing occurred for each model: filtered and unfiltered. Unfiltered data was used simply to determine if any noise artifacts from the resampling of the experimental data were present, and if they made a difference in the correlation analysis. Filtered calculation used a 3rd order type II Chebyshev filter to insure a rapid rolloff at a cutoff frequency of 2 kHz, and linear phase throughout the passband region. The 2 kHz cuttoff frequency was chosen because the simulation of model 8 could only be accomplished from 500 to 2000 Hz in linear steps of 500 Hz without crashing the program. In all cases the correlation of both of the binaural channels with the experimental data was calculated, and then a correlation of the superposition of the two binaural channels with the experimental data was done. Superposition of the binaural impulse response data was done because the experimental impulses are monaural, and at any point in space, the sound field will be a summation of incoming waves from both the left and right hemispheres.
As can be seen from the data included in the appendices, the unfiltered data did not correlate well with model data, exhibiting normalized correlations from approximately 0.25 to 0.3 for the superimposed impulses. This low number should be expected because the experimental data contains a full 44.1 kHz bandwidth, plus noise artifacts.
Filtered data yielded results which could be considered not exactly great, but more in favor of the RAYNOISE program. Superimposed binaural impulse responses correlated much better this time, with model 8 (the most complex model) having a maximum normalized correlation of approximately 0.6. The other models had similar maximum correlations of approximately 0.5 to 0.55.
So what can be said about this analysis? The filtered data should be rated as a higher standard by which to gauge the accuracy of RAYNOISE because it is band-limited to the frequencies at which RAYNOISE did the analysis. Although the correlation did not come out to be close to one, it seems to validate RAYNOISE as an accurate program considering the following: 1) the location of the experimental microphone was measured only approximately, and indicated in the report on a grid with estimated resolution of 2 feet per element. Therefore the virtual microphone may in fact be located several feet from the location in the hall at which the experimental data was recorded. 2) Due to memory constraints and substandard computational performance, RAYNOISE was only able to use a limited number of rays to generate these results over a narrow band of frequencies.
With regard to the complexity of the model none of the results seem to justify having a complex rendering of the room to obtain meaningful results via impulse response calculations. As we will see in the acoustical analysis section, this is not the case when calculating the acoustical parameters of the room. Perhaps there is some type of flaw or over-simplification in the RAYNOISE impulse response calculation algorithms, or perhaps the room materials are more important when generating the impulse responses.
Does this infer that possibly RAYNOISE is an inferior program for room analysis? Results indicate that this is not necessarily true. Consider that even if the program is not 100% correct, if its algorithms are based on the laws of physics (and I'm sure that they are), that any corrections to the room that improve its acoustics devised through the use of the program will likely improve the acoustics of the real room. I would deem the correlation analysis inconclusive and believe that ultimate judgement of this should be withheld until someone can get RAYNOISE to run this type of analysis on a realistic model at higher frequencies without crashing.
ACS597E students made the following general determinations about the Recital Hall Acoustics from subjective surveys and experimental analysis:
Calculations with RAYNOISE do in fact confirm these results, and offer possible explanations and acoustical corrections.
Eight acoustical parameters at 1 kHz were chosen to analyze to determine how the architectural features of the room influence it's acoustics.
Graphics of the acoustical analysis data can be found at the URL: http://www.acs.psu.edu/users/smithsh/hall_acoustics.html
Following is a list of these parameters, their definitions, and what one can conclude about the acoustics of the room based on the data. The benefit of running calculations on a series of increasingly complex models is clearly shown in the data, as one can determine which architectural features are responsible for the acoustical behavior of the room as each of the features is "turned on"
SPL Sound Pressure Level A stationary measure of the pressure in dB ref 20 micro-Pascal.
The benefits of the wedge structures of designed into the back wall are evident when considering SPL distributions throughout the room. From examination of the data, the SPL levels become drastically more uniform in models 5-8 than they were in any of the previous models. Although the room tends to continue exhibiting regions of elevated sound pressure, the wedges serve to break up constructive and destructive interference in a manner that no other element in the room can duplicate.
STI Speech Transmission Index The normalized value of averaged broad-band Modulated Transfer Functions in a Room. A Modulated Transfer Function is a response curve describing attenuation of a frequency band by the room. STI of 1 indicates excellent speech intelligibility. A variant of STI is RASTI which only uses values from the 500 and 2k Hz frequency bands for a similar calculation.
Again the wedges in the back wall demonstrate their worth, improving the intelligibility of speech by breaking up interference patterns in the room. Prior to their introduction, the clearest speech could only be obtained only in the rear corners or along the walls. Although the speech intelligibility of the modeled room is not that good - typically 0.4 to 0.5 with 1 being excellent, 0.8 to 1 being good - it would be horrendous without the wedges. Presumably the room is never used for any public speaking, or that the sound reinforcement speaker overhead accommodates the need for improved STI. It is interesting to note that the organ enclosure does slightly reduce the speech intelligibility on the stage-left side of the hall, while slightly improving it in the rear, stage-right corner.
C80 Clairity: dB Ratio of early to late energy. Measures how well defined music content is. Low clarity is associated with poor definition of individual musical sounds.
From examination of clarity as a function of model number it is evident that the rear wall of triangular pyramids (model 5) has a good deal to do with improving clarity in for most of the room. It not only evens out the distribution of clarity, but increases the levels. Likewise, the overhead stage reflector is instrumental in improving the clarity for listeners in the front rows.
Again, the organ enclosure serves to interfere with the sound field in such a manner as to elevate the clarity in the rear stage-right corner of the hall and along the right wall, while reducing the clarity for audience members in the font, stage-left rows.
EC Echo Criterion A quotient which, if exceeding 1.8 for musical performance, will result in 50% of the audience hearing an echo.where
It is interesting to note that the features of the hall go a long way in suppressing echos as they are "turned on." Hot spots in these graphics indicate regions in which members of the audience are likely to experience echo problems. One of the features that goes a long way in breaking up echo hot spots in the room is the overhead reflector. Although regions of the audience still tend to exhibit areas whit elevated likelyhood of echos, a real problem occurs at the hot spot in the first few rows. This could most likely be corrected by installing hanging reflectors called "clouds" near the front of the hall. Not only will they improve the early reflections for these rows, but they should obstruct the high amplitude sound off the reflector which is most likely causing this problem. It is likely that the back wall is contributing to the echo hot-spots in front of the stage in models lacking the reflector, but this problem is solved by applying a curtain to the back wall. The reflector can be singled out as the source of the elevated levels by comparing the colormap scales on the right of the EC plots with the data in front of the stage for models 6 and 7.
LE Lateral Efficiency: Measures impression of spaciousness. A ratio of early to late arriving reflections incorporating an angular measurement from the "ear-to-ear axis" Small Lateral Efficiency indicates the need for increased early reflections.
One of the most startling revelations of the room's acoustical behavior can be seen in the Lateral Efficiency data, and confirms observations by the ACS597E group: the room suffers from a severe lack of lateral reflections. All of the hall models, including the most complex model demonstrate a very low degree of Lateral Efficiency compared to the first two rows, which receive lateral reflections from the angled stage walls. This validates the suggestions that some type of acoustical diffusers be placed along the side walls to improve lateral reflections, resulting in a greated feeling of spaciousness.
ERR Early Reflection Ratio: Ratio of early sound energy to direct sound energy within 50 ms. A low ERR indicates the need for more early reflections.
Just as telling at the Lateral Efficiency Data, this demonstrates another serious flaw in the room. For each model from the simple box up to the complex rendering of the hall, approximately the first third of the hall never receives any improvement in early reflections. This should be corrected because it serves to narrow the apparent source width, and reduces the quality of the sound.
RT60 Reverberation Time: Time for sound level to decay to -60 dB.
The ACS597E TEF MLS analysis of the room arrived at the following reverberation times:
Octave Band Frequency (Hz) 63 125 250 500 1000 2000 4000 8000 Location Row 9 or 10 2.35 2.06 1.58 1.85 1.89 1.93 1.55 1.15 Row 15 2.25 2.18 1.78 1.86 1.81 1.75 1.53 1.05 Row 3 1.5 1.77 1.51 1.69 2.06 1.87 1.61 0.99
It can be seen from the graphics of reverberation times that the calculated results compare quite favorably to those measured in the real hall, with the largest deviations being on the order of 0.1 seconds. Schwenke et. al. mention some concerns with this reverberation, indicating that it is too low for an empty hall. I would tend to agree. Most musical performance should occur in a room that possesses a reverberation time equal to or exceeding 1.7 seconds.
Schwenke et. al. believe that the reverberation time will drop a full second with an audience in the room, and that will have a drastic effect on the acoustics. With respect to the simulation of the room, I would agree. The audience areas in the simulation were constructed of truncated rectangular sections with the top and front elements having the absorption coefficients of medium upholstered seats. The sides and back were modeled by setting their absorption coefficients equal to steel to simulate the seat backs, sides, and pans. It would seem fortuitous that the model agrees so well with the measurements, and it would certainly be not only interesting, but possible to see what would happen if the model were altered to simulate an audience with the appropriate absorption coefficients.
I-DEAS and RAYNOISE are computer software packages that can be used to model a room to a fairly good degree of certainty. Although correlation data of echograms generated from experimentally measured data in the actual room with model room impulses was fairly indifferent, observations of acoustical parameter calculations of the room generated by RAYNOISE confirmed previous experimental results, and have provided some additional insights into the behavior of the hall. From the combined correlation and acoustical data it can also be concluded that a complex model is not necessary to gain good insight into the behavior of the room as long as one makes sure to include features up to a certain level of detail - at least those which may tend to be a dominant feature in the room.
Initially, is was believed that the room simply suffered from lack of lateral reflections. Not only was this confirmed, but also this analysis has shown the room also has a problem with echos in the first rows of the audience caused by the stage reflector. Additionally, this study confirmed that the organ enclosure does have an effect on the acoustics of the hall, causing left-right imbalances in speech intelligibility and clarity. This study clearly showed that the dominant feature of the room is the wedges built into the back wall. Without this feature the acoustics of the room would be drastically imbalanced, resulting in "hot spots" of good and bad acoustics.
Although time constraints of this project did not allow for the inclusion of acoustical correction to the room it is quite obvious that one would start implementing corrections by designing diffusing surfaces for the side walls to improve the lateral efficiency and early reflections, as well as including some type of hanging reflectors to break up strong signals off the overhead stage reflector.
On an interesting note, the good agreement between the actual and model reverberation times indicate that the chosen method of modeling the seating area by applying upholstery absorption to the top and front of the seating constructs, and steel absorption coefficients to the remaining audience construct is fairly accurate. In most cases, a clear method for modeling such constructs has not been defined, and they remain difficult to approximate.
Suggestions for future work include:
In summary, with the exception of implementing acoustical changes, all the objectives of this project were achieved. As a final statement, it is my personal belief that for future analyses it would make a good deal of sense to obtain a CAD package such as AUTOCAD to cut down on the drastic amount of time required to construct room models.
Paper, computer models, data, graphics Copyright 1998, Steven Smith