Conference PaperPDF Available

From the sound up: Reverse-engineering room shapes from acoustic signatures


Abstract and Figures

Typically, architects and acousticians design rooms for music starting from a model room shape known from past experience to perform well acoustically. We reverse the typical design process by using a model sound signature to generate room shapes. Our method builds off previous research on reconstructing room shapes from recorded impulse responses, but takes an instrumental, design-oriented approach. We demonstrate how an abstract sound signature constructed in a hybrid image source-statistical acoustical simulator can be translated into a room shape with the aid of a parametric design interface. As a proof of concept, we present a study in which we generated a series of room shapes from the same sound signature, analyzed them with commercially available room acoustic software, and found objective parameters for comparable receiver positions between shapes to be within just-noticeable-difference ranges of each other.
Content may be subject to copyright.
Proceedings of the Institute of Acoustics
Vol. 37. Pt.3 2015
Willem Boning Arup Acoustics, New York, NY, USA
Alban Bassuet Tippet Rise Art Center, Fishtail, MT, USA
The process of designing a room for music often begins with a rough shape known to perform well
acoustically and functionally. The shape is then refined through analysis, simulation and
optimization to meet acoustical targets. But what if it were possible to invert the process, starting by
composing the room’s sound and then generating a matching shape? In this paper, we propose a
method for acoustical reverse-engineering. We describe techniques that allow the designer to
construct a sound signature characterizing the room’s acoustics and use that signature to generate
a solution space of corresponding room geometries.
Acoustical reverse-engineering can be considered a design-oriented inverse problem. When the
parameters that define a model or system are unknown (like the shape of a room), an inverse
problem can be used to reconstruct those parameters from the system’s output data (like acoustical
measurements). Perhaps the best-known example of an inverse problem related to acoustics is the
drum shape problem posed by Kac, who claimed that it is possible to infer the shape of a drumhead
from the sound it makes when struck.1 Gordon and Webb later found that a single sound impulse
could in fact describe multiple, isospectral drumhead shapes (fig. 2).2 Their finding reflects a
structural difference between forward and inverse problems: a forward problem has a single
solution, whereas an inverse problem may consist of a model space of multiple solutions that are
either equivalent or probabilistically distributed.3 When a drumhead is struck, a single, unambiguous
impulse is produced. That same impulse, however, could have been produced by different
drumhead shapes.
A simple inverse problem can be used in the early stages of design to determine an acoustically
appropriate room volume. The volume can be back-calculated by defining the seating capacity and
desired reverberation time and rearranging the variables in Sabine’s formula.4 If reverberation time
were considered the main driver of acoustic quality, solving this inverse problem would suffice as a
design template. But early reflections, which are essential for strength, clarity, intimacy and
envelopment, are just as important to the listening experience.5,6,7 A room reverse-engineered from
reverberation time and seat count will not produce a particular shapeany convex, non-coupled
geometry will do as long as it conforms to the required volume. Early reflections, however, are not
Proceedings of the Institute of Acoustics
Vol. 37. Pt.3 2015
Figure 1: Gordon and Webb pose with paper models of isospectral drums.
shape-agnostic. Two rooms with the same reverberation time, one tall and narrow and the other
short and wide, will sound different because early reflections will arrive at listeners’ ears from
different directions, at different delays and at different levels of attenuation. Acknowledging the
importance of early sound requires a more comprehensive approach to acoustical reverse-
engineering, one with early reflections at the heart of the process.
Our method for acoustical reverse-engineering draws from two lines of room acoustics research.
The first, acoustic scene reconstruction, involves using inverse methods to infer deterministic room
shapes from early reflections. Gunel describes a technique for estimating a physical room's shape
by analyzing a directional impulse response.8 Antonacci et al. describe a method for inferring a two-
dimensional room boundary by analyzing a series of mono-channel impulse response recorded at
different locations.9 Dokmanić, Lu and Vetterli reconstruct a two-dimensional room shape from an
impulse recorded on a single mono microphone10 and Dokmanić et al. extend the method to a
three-dimensional room by analyzing an impulse response recorded by an array of arbitrarily placed
mono microphones.11 This line of research is forensic in its pursuit of a single solution, with
advances made as increasingly simple data is used to infer increasingly complex room shapes.
Rather than generating a single shape from an impulse response, our aim is to create a solution
space offering acoustically equivalent but geometrically diverse outcomes to the designer.
We also draw from work done by Bassuet and Woodger to patent a process for designing rooms
based on virtual acoustic signatures.12 The authors synthesized a number of ideal signatures based
on impulse responses recorded in a survey of historical rooms for music. They then developed a
room in pursuit of each signature, starting with a basic shape and refining it iteratively until acoustic
analysis software confirmed a match. The reverse-engineering method we propose in this paper
begins with a virtual sound signature similar to that described by Bassuet and Woodger but
presents a new technique for translating the signature into a room shape. Rather than using
intuition and iteration as the mechanisms for design, and rather than requiring the designer to hone
in on one solution, our method automatically generates a geometric solution space that allows the
designer to freely explore a range of unforeseen shapes and forms.
Proceedings of the Institute of Acoustics
Vol. 37. Pt.3 2015
Our method for acoustical reverse-engineering begins with a sound signature that characterizes the
acoustical environment of an audience area in a future room. The signature will be used to infer the
paths sound waves must take to produce the same acoustic impression, and those paths will in turn
be used to generate a series of sound-reflecting surfaces that define the room's overall shape.
The sound signature we adopt for the purpose of reverse-engineering is a "hybrid model" signature
consisting of the sound source, a series of image sources and statistically-defined reverberation
(fig. 2). The hybrid model was first described by Vorländer13 has served as the basis of a number of
room acoustic simulation softwares, including Odeon,14 CATT,15 Spat16 and RAVEN.17 The model
cannot represent all of the information that would be captured in a recorded impulse response and it
is poor at characterizing rooms in which the sound impression is dominated by properties of
diffraction, modal behavior, high absorption, and (if a statistical tail is used) non-ergodic
reverberation.18,19 But it has proven effective for characterizing medium to large Sabinian rooms
with low absorption and strong, relatively broadband early reflections, all qualities typically
considered to be desirable in rooms for unamplified music. And crucially, the hybrid model contains
enough spatial information to construct a corresponding room geometry.
Source # Distance (m) Atten. (dB)
Image Source # Az. angle (deg) El. angle (deg) Delay (ms) Dist. atten. (dB) Add. atten. (dB)
1 37.6 2.3 11.80 -26.9 0.0
2 324.6 2.2 15.30 -27.3 0.0
3 46.5 38.7 42.40 -30.2 0.0
4 316.6 37.2 45.70 -30.5 0.0
5 0 60.7 57.90 -31.6 0.0
6 168.4 1.1 96.20 -34.1 0.0
7 200 1.1 97.80 -34.2 0.0
8 37.6 55.7 64.10 -32.0 0.0
9 324.6 54.4 66.10 -32.2 0.0
10 163.3 24.4 111.00 -35.0 0.0
11 204.7 21.2 113.30 -35.1 0.0
Reverberation # Reverb. time (s)
Figure 2: Sound signature displayed in table format.
For purposes of acoustical simulation, the image sources in the sound signature are defined in
terms of incidence angle, delay time, attenuation due to distance and attenuation due to absorption
or scattering. For purposes of inferring their reflection paths in space, delay is converted into
distance (according to the speed of sound in air) and the images are located as points in space
relative to a single receiver. To extend an acoustical signature beyond one "originating" receiver
location, the image sources must be made audible to multiple receivers across an audience area.
But do the images translate relative to each receiver or do they remain fixed in place with reference
to the originating receiver? If we were to take the first approach, the incidence, delay and
attenuation of each image would remain constant, guaranteeing uniform sound at every seat in the
house. While this scenario may seem like the democratic ideal, it is nevertheless impossible to
achieve physically, as each set of images for each receiver would require its own set of reflection
surfacessix images multiplied by 500 audience members would yield 3,000 surfaces, many of
which would occlude each other (fig. 3). We take the second approach, in which one fixed set of
image sources, and one set of corresponding surfaces, serves the entire audience area. All of the
images remain audible from seat to seat but gradually change in incidence angle, delay and
attenuation moving away from the originating receiver. The range of perceived incidence for each
image source can be found by translating the audience perimeter along a vector from the originating
receiver to the image source as shown in figure 4. The acoustic variability can thus be predicted
and controlled to ensure that the acoustic character remains consistent across the audience area.
Proceedings of the Institute of Acoustics
Vol. 37. Pt.3 2015
Figure 3: Image sources referenced to one originating receiver (l) and translated to give the same
early sound impression to multiple receivers (r).
Figure 4: Image sources fixed in space for all receivers (l) and the resulting variability of perceived
image source incidence across the receiver area.
2.1 Sound Signature Design Interface
A sound signature may be abstracted from an impulse response as shown by Antonacci20 or
derived from a model of a room by image source method (ISM) as described by Allen and Berkley 21
and Borish.22 Alternately, a sound signature can be designed from scratch, offering the designer the
opportunity to invent an entirely new acoustical environment. To facilitate the design of new
acoustic signatures, we created a software tool that allows the user to construct, modify and listen
to sound signatures in real time. In the tool’s user interface (UI), constructed in Processing, the
designer can set his or her distance from the virtual sound source, position and reposition image
sources, and adjust the reverberation time (fig. 6).
As the user adjusts the acoustical signature, its parameters are ported from the UI to an audio
simulation patch constructed in MaxMSP with Spat objects. The patch handles the direct sound and
image sources by panning, delaying and attenuating an anechoic music signal through individual
taps, and simulates reverberation using Spat's cluster and reverberation objects, with the room's
mixing time estimated based on the image source delays. All of the output taps are then combined
and decoded for second-order Ambisonics or binaural playback. Because updates to the sound
signature are processed and auralized in real time, the user has the choice of listening to changes
while making adjustments or can save and retrieve sound signatures to listen to A-B comparisons.
The user is also free to switch between different anechoic samples to test how well the sound
signature supports different types of music.
The sound signature UI is currently limited by the fact that it is reality-agnostic, in that the user can
set any image sources and reverberation he or she wants regardless of whether they are physically
realizable. Using the interface thus requires some forethought about the future room's overall
envelope and audience capacity. For example, two 15 ms-delayed images arriving from either side
Proceedings of the Institute of Acoustics
Vol. 37. Pt.3 2015
Figure 5: Sound signature design interface.
of the room would likely be very difficult to achieve in a 2,500-capacity concert hall (unless it were
extremely narrow and extremely long). Likewise, an image source located straight down would be
hard to realize in any room. To increase the likelihood of a realizable sound signature, the user can
derive image sources from the future room's maximum envelope and use those to get a rough
sense of scale.
As the sound signature produced in the UI will not remain constant from seat to seat in the eventual
room, it is the designer’s task to craft a signature that can be successfully extended over a larger
audience area. To experience how the signature will vary, the interface allows the user to translate
his or her listening location away from the originating receiver position to hear the source and image
sources from different vantage points. While a single sound signature can be applied to the entire
audience area, it may also be desirable to construct multiple signatures and assign them to different
audience areas within the same room. The designer can use multiple signatures to create a room
with distinctly different acoustical environments or, alternately, to head off acoustical defects. For
example, an image source that is well-integrated at the originating receiver position may arrive
outside the window of integration at another receiver position some distance away, requiring the
signature to be modified or replaced.
Proceedings of the Institute of Acoustics
Vol. 37. Pt.3 2015
The image sources in a sound signature are made audible in a room by deducing the location,
orientation and extents of planar, sound-reflecting surfaces. A single order of reflection yields one
deterministic solution. The first-order reflection plane and reflection point for the originating receiver
can be found by inverting the image source method as shown in figure 6. At two orders of reflection,
however, a large solution space opens up. The first reflection point (with respect to the receiver)
may be located anywhere along the incidence vector up to the plane of first-order reflection. The
second reflection point can then be located at any point on the surface of an ellipsoid with the first
reflection point and source as focii. Figure 7 shows a partial second-order solution space.
Figure 6: First-order reflection solution.
A reflection surface or pair of surfaces must be extended to make an image source audible to a
larger audience area. This is done by inverting the ISM visibility test for a sample of receivers on the
audience area boundary, as shown in figure 8. Surfaces can also be extended by the same inverse
test to make images of a larger source area audible. To ensure broadband reflections reach the
receivers at the boundary of the audience area, the surface edges must be further offset. The offset
distance can be approximated by applying Rindel's equation for sizing reflectors, which factors in
reflection angle and characteristic distance between source, reflection point and receiver.23
The materiality of a surface is determined by the originating image source's attenuation. If its
attenuation is due to distance alone, a rigid, massive and smooth material will be required. If
additional attenuation is specified, the material must be assigned scattering or absorbing properties.
Proceedings of the Institute of Acoustics
Vol. 37. Pt.3 2015
Figure 7: Second-order partial solution space with sample paths drawn (above) and selected path
Proceedings of the Institute of Acoustics
Vol. 37. Pt.3 2015
Figure 8: Second-order solution extended to multiple receivers and sources.
3.1 Solution Space UI
We constructed a user interface in Grasshopper, a parametric modeling plugin for Rhino, that
allows the designer to navigate the solution space and create reflection geometry for each image
source in a sound signature (fig. 9). The user can toggle between first and second-order solution
modes. In first-order mode, only one solution is possible and no further input is required from the
user: what you see is what you get. In second-order mode, the user is given controls to explore the
larger solution space: a one-dimensional slider sets the position of the first reflection point along the
incidence vector and a two-dimensional slider sets the second reflection point's UV coordinates on
the ellipsoid. Given the two reflection points, the patch calculates the angles of the reflecting
surfaces and their extents for the audience and source areas in the 3D model. As the user adjusts
the points, he or she can observe the reflecting surfaces shift and morph in real time, and is free to
pick an outcome based on aesthetic, functional or other considerations with the confidence that
every solution will be acoustically equivalent.
The UI allows the user to visualize solutions for all of the image in the signature simultaneously.
Viewing the individual solutions together helps the designer position surfaces so that they do not
collide with each other or occlude each others' reflection paths. To verify that no surfaces are
obstructing the image sources' audibility, the user can toggle on a forward ISM visibility check to
identify any "deaf spots" in the audience. The same check can be used to identify any unintended
early reflection paths, for example a first-order reflection off one of the two surfaces designed to
realize a second-order solution.
Proceedings of the Institute of Acoustics
Vol. 37. Pt.3 2015
Figure 9: Solution space explorer UI.
The group of reflection surfaces that emerge from the solution space UI will be discontinuous, so
the designer must complete the room’s shape by adding infill geometry. Once the room is closed,
the patch calculates the room's volume and calculates the Sabins required to meet the sound
signature's reverberation time, which the user may then apply to the audience area and/or the infill
surfaces. (If any image sources are attenuated more than due to distance alone, some of the
Sabins will be automatically applied to those images' reflecting surfaces.)
Once a room geometry is completed, it must be analyzed to determine whether its acoustics match
those constructed in the sound signature design interface. Strictly speaking, there should be no
audible difference between the virtual sound signature and an impulse response recorded at the
originating receiver position in the resulting physical room. The same expectation of similarity
applies to displacements within the virtual sound signature and their corresponding positions in the
real room.
Because we have not yet completed a project with reverse-engineering as the basis of design (two
are currently in development), we present a partial validation as proof of concept. We derived a
sound signature from an existing room shape and used it to reverse-engineering two radically
different room shapes. Would the three rooms sound the same? We simulated their acoustics in
CATT v9 and analyzed the results by comparing numerical measures, visualizing 3D Impulse
Responses (3DIRs), and conducting A-B listening tests. In addition to checking for similarity
between comparable receiver positions between the three rooms, we also wanted to assess the
variance in sound impression within each room. Would generating the room from a single set of
fixed image sources produce an acoustical pattern that would hold over a large audience area?
4.1 Setup
We derived an initial sound signature from a design for a long and narrow hall with a cross-shaped
section. The room had been designed to have reverberant yet intimate and enveloping sound
characteristics, supported by an array of strong early reflections from multiple directions. We
estimated the reverberation time by Sabine formula and identified key early reflections by ISM for a
source on stage and a receiver placed a little over halfway back in the audience area. The resulting
sound signature is shown in figure 3 above. Using the solution space UI, we then constructed
Proceedings of the Institute of Acoustics
Vol. 37. Pt.3 2015
Figure 10: Three room shapes created from one sound signature.
surfaces corresponding to each image source in the signature, sizing them to reflect sound from the
source to the entire audience area. Where surfaces unavoidably overlapped or intersected, we
trimmed them to ensure a continuous, occlusion-free envelope. We then added infill geometry to
complete the room shape. We carried out this process three times, creating three different room
shapes (fig. 10-11). Room A is a recreation of the originating room’s cross-shaped geometry. Room
B is a product of first-order solutions only, resulting in a symmetrical but non-orthogonal form. We
constructed Room C from a combination of first and second-order solutions, enabling us to craft a
more organic, asymmetrical shape.
Figure 11: Interior views of rooms A, B and C
Proceedings of the Institute of Acoustics
Vol. 37. Pt.3 2015
4.2 Computer simulation
To simulate the acoustics of each room, we imported their geometries into CATT v9. We assigned
the reflecting, infill and stage surfaces a mid-frequency absorption coefficient of 0.02 and a
scattering coefficient 0.3 and assigned the audience plane a mid-frequency absorption coefficient of
0.98 and a scattering coefficient of 0.9. (We assigned greater-than-typical values for absorption in
order to produce a clearer comparison of the effects the rooms’ reflecting surfaces). We specified a
single, omnidirectional source at the same location as in our initial ISM derivation and in addition to
the originating receiver, added 13 other receiver positions spread across the audience area. Each
room configuration was rendered in CATT-TUCT v1.1 using algorithm 2 with 1,000,000 rays emitted
and diffraction disabled.
4.3 Results and analysis
Figure 12 shows values for ISO-3328:3–2012 parameters G, T-30, EDT, C-80, and LF, extracted
directly from the CATT-TUCT echogram at the 500 Hz octave band, with the range of just-
noticeable-difference (JND) for each parameter indicated by error bars. The results show very
strong acoustical similarity between the three rooms, with nearly all parameter values falling within
JND ranges of each other from position to position. The results also show evidence of a relatively
consistent sound impression across the audience area (save for receivers at the very front, where
direct sound inevitably dominates, and receiver 14 at the very back). The most notable
discrepancies are that room B’s EDT values are consistently lower and room B’s LF values are
consistently higher than those of room A. The differences are slight, however, with almost all values
falling inside the JND range.
Figure 13 shows 3DIRs for receiver positions 6, 12 and 14. The 3DIRs were created by spatialzing
2nd-order B-format impulse responses generated in CATT according to a method developed by
Bassuet,7 and show the directionality, intensity and arrival time of sound reflections relative to the
direct sound. Like the numerical parameters, the 3DIRs indicate strong similarity between
comparable receiver positions. The delay time window and directionality of reflections is consistent
between rooms with only slight differences in energy level. The only notable discrepancy is at
receiver positions 12 and 14, where a reflection arriving from the upper left is visible in rooms B and
C but not in room A. Within each room, the reflections change in intensity, delay and incidence
angle from position to position but importantly, they all remain present in the signature, again
confirming the extension of the originating signature into a consistent pattern across the audience
As a final subjective test of whether the rooms would sound the same, we carried out a blind A-B
listening survey in the Arup SoundLab, focusing on receiver positions 6, 9, 12 and 14. 12
participants, all acoustic and audiovisual consultants, were asked to compare 11 pairs of 2nd-order
B-format impulse responses convolved with an anechoic sample of Handel’s Water Music and
decoded for Ambisonics playback over the SoundLab’s speaker array in terms of loudness, clarity,
early-to-late energy balance, image width, envelopment and reverberation. For each parameter,
participants could respond that sample B was greater than A (+1), equal to A (0), or less than A (-1).
Participants were also asked to note the most significant difference they heard between the two IRs.
For each comparison, the sample was first played twice, switching from A to B and then from B to A
at the halfway points. Participants were then allowed to request as many repetitions of the sample
in A-B and/or B-A order as they needed to complete the comparison. Each comparison included
one IR from room A and one IR from either rooms B or C, with the playback order varied. The one
exception was one pair of identical IRs, included as a control for differences between the two halves
of the anechoic sample.
Proceedings of the Institute of Acoustics
Vol. 37. Pt.3 2015
Figure 12: Comparison of acoustical parameters for the three rooms. Position 8 is the originating
Figure 14 shows averages of perceived differences in acoustical parameters in rooms B and C
compared to room A for receiver positions 9, 6, 12 and 14, for an average of the positions and for
the control. A score of +1 would indicate a unanimous perception of a parameter being greater in
room B or C than in room A. A score of -1 would indicate unanimous perception of the parameter
being less in room B or C than in room A. A score of 0 indicates either unanimous perception that
the parameter was the same in both rooms or evenly-balanced disagreement. The scores are
generally low, and their absolute values are for the most part less than those of the same-IR control,
indicating that the rooms sounded the same or at least very similar depending on the participant
and the receiver position. Written comments support the same conclusion. For the comparison
between room C and A at receiver position 6, for example, participants wrote comments including,
“Very subtle,” “Hard to compare,” “Didn’t hear much difference” and “Any difference too subtle to
Proceedings of the Institute of Acoustics
Vol. 37. Pt.3 2015
Figure 13: 3D Impulse responses at receiver positions 6, 12 and 14.
Figure 14: Averages of perceived differences in acoustical parameters in rooms B and C compared
to room A.
Proceedings of the Institute of Acoustics
Vol. 37. Pt.3 2015
distinguish.” Averaging the perceived differences for each parameter over the four receiver
positions, the only distinction between room B and room A seems to be that it is slightly more
enveloping when measured against the control. None of the averaged parameter differences in
room C exceed those of the control by more than 0.1, making it difficult to claim any overall
From the results of our numerical, visual and aural analyses, it is clear that the three rooms we
generated from a single sound signature sound nearly the same, and from some receiver positions
are acoustically indistinguishable. The only consistent differences we observed across the three
analyses involve spatial impression. We attribute these differences not to a fault in our reverse-
engineering method but to two aspects of our validation procedure that could have been better
controlled. First, the relatively high scattering coefficient we applied to the reflecting surfaces likely
attenuated second-order reflections more than first-order reflections, which may explain why room
B, reverse-engineered to return first-order reflections only, was perceived to more enveloping.
Some variance in spatial impression between the halls may also be due to small “deaf spots”
caused by trimmed reflection surfaces to make them fit together into a continuous envelope. Room
A required substantially more trimming than rooms B and C in order to match the originating room’s
shape. A forward ISM analysis of selected surfaces in the three rooms demonstrates that in room A,
upper right side reflections are indeed inaudible from the right edge of the audience area (fig. 15).
Figure 15: Upper right side reflections in rooms A, B and C. Note the lack of reflections reaching
receivers next to the right side wall in room A.
In this paper we presented a new method for acoustical reverse-engineering. We illustrated how a
virtual sound signature can be used to generate a solution space of geometries to achieve the same
acoustic impression physically. We also showed that a single signature can be used to define an
acoustical environment with an acceptable range of variation for a larger audience. In our validation
experiment, we demonstrated that the solution space for one sound signature can produce room
shapes that look different but sound the same, not just from one listening position but across the
entire audience area.
By linking a sound signature to all of its possible geometric solutions, reverse-engineering frees the
designer to experiment outside the aural, formal and typological conventions of concert hall design.
The designer can sculpt any sound knowing it will be physically realizable and can craft any shape
knowing it will not compromise the acoustics.
From an aural perspective, reverse-engineering offers an opportunity to explore new sound
aesthetics that might be unlike those of any existing room for music. Conversely, it offers an
Proceedings of the Institute of Acoustics
Vol. 37. Pt.3 2015
Figure 16: The interior of room B showing basic geometry and added architectural flourishes.
opportunity to replicate the sound from an existing hall without needing to recreate its shape. A
room can sound like a shoebox without looking like a shoebox, for example. Our method also gives
the designer control over how uniform or disparate the sound impressions are within a single space.
A reverse-engineered hall could offer listeners the choice between two (or more) distinct but equally
engaging acoustical environments. Control over the consistency of a room’s sound sets reverse-
engineering apart from real-time auralization techniques, which allow the designer to modify a room
shape and hear the resulting sound in real-time, but only from one listening position at a time, and
with no guarantee that the sound impression will extend to a larger audience area.
While reverse-engineering gives the designer a great amount of sonic flexibility, it currently limits
control over the shape of the late sound to setting reverberation time. Bradley and Soulodre,
Bassuet and Lokki have found that variations in late energy can affect perceptions of envelopment
and spaciousness,24,7,25 but the relationship between geometry and late sound is still relatively
unexplored, so more research must be done before it can be integrated into the reverse-
engineering process.
From a visual and formal perspective, reverse-engineering offers an opportunity to create new
spatial environments that are not simply variations on the traditional concert hall models. In our
validation exercise, we took advantage of this opportunity to create three very different looking room
Figure 17: Diagram for a new performing arts venue in Montana featuring a reverse-engineered
inner shell.
Proceedings of the Institute of Acoustics
Vol. 37. Pt.3 2015
shapes from the same initial sound signature. The freedom comes at the relatively small price of
refraining from napkin sketches at the start of a project in favor of developing a flexible design
language capable of transforming the raw geometric output of the solution space. It goes without
saying that the final rooms do not need to be hard-edged and gray-colored (fig. 16). Above, we
mentioned the possibility for a sound signature from a historical room to be reverse-engineered into
a radical new shape. By the same token an architect should also be able to take a radical new
sound signature and reverse-engineer it into a conservative, orthogonal room shape. We also note
that surfaces generated through reverse-engineering do not need to be connected into a continuous
room shape. For a new performing arts center in Montana, we are proposing an inner array of
reverse-engineered panels open to a larger acoustical volume (fig. 17). In this “deconstructed”
concert hall, the panels will realize the images in our sound signature by returning early reflections
to the audience and the outer box will provide the reverberation.
Acoustical reverse-engineering cannot realize all types of room sounds, nor is it capable of
generating all types of room shape. A sound signature dominated by uneven late sound, for
example, would be hard to realize, and the solution space will never automatically return a curved
surface. But despite its exceptions, reverse-engineering remains a powerful tool for exploring new
sonic and formal environments. By freeing the designer from preconceptions about how a room for
music should look or sound, we hope the method will stimulate designers to invent a great diversity
of engaging, surprising and stimulating spaces for performing and listening to music.
The work presented in this paper was initiated and developed in partnership with Matthew Berstch.
The authors would also like to acknowledge the generous technical assistance of Yshai
Yudekowitz, Charles Avis, Han Dong and Terence Caulkins.
1. M. Kac. 'Can one hear the shape of a drum?', The American Mathematical Monthy 73(4), 1-
23. (1966).
2. Gordon and D. Webb. 'You can't hear the shape of a drum', American Scientist 84(1) 46-55.
3. A. Tarantola. Inverse Problem Theory and Methods for Model Parameter Estimation.
Philadelphia: Society for Industrial and Applied Mathematics. (2005).
4. L. Beranek. Concert Halls and Opera Houses, 2nd ed. New York: Springer. (2004), p. 541.
5. A.H. Marshall. 'Acoustical determinants for the architectural design of concert halls',
Architectural Science Review 11(3), 81-87. (1968).
6. Y. Jurkiewicz and E. Kahle. 'Early reflection surfaces in Concert Halls - a new quantitative
criterion', Proc. Acoustics '08. Paris (2008).
7. A. Bassuet. 'New acoustical parameters and visualization techniques to analyze the spatial
distribution of sound in music spaces.' Proc. International Symposium on Room Acoustics.
Melbourne (2010).
8. B. Gunel. 'Room shape and size estimation using directional impulse response
measurements', Proc. Forum Acusticum. Seville (2002).
9. F. Antonacci, et al. 'Inference of Room Geometry From Acoustic Impulse Responses', IEEE
Transactions on Audio, Speech and Language Processing 20(10). (2012).
Proceedings of the Institute of Acoustics
Vol. 37. Pt.3 2015
10. I. Dokmanić, Y.M. Lu, and M. Vetterli. 'Can one hear the shape of a room: the 2-D
polygonal case', Proc. ICASSP. Prague (2011).
11. I. Dokmanić, et al. 'Acoustic echoes reveal room shape', Proc. National Academy of
Sciences of the United States of America. (2013).
12. G.A. KnicKrehm, A. Bassuet, G. Ellerington, and A. N. Woodger. Methods and systems for
improved acoustic environment characterization, U.S. Patent No. 8,396,226. (2013), sheet
13. M. Vorländer. 'Simulation of the transient and steady-state sound propagation in rooms
using a new combined ray-tracing/image-source algorithm', J. Acoust. Soc. Am. 86 (1), 172-
178. (1989).
14. G. Naylor. 'Treatment of Early and Late Reflections in a Hybrid Computer Model for Room
Acoustics', Proc. 124th ASA meeting. New Orleans (1992).
15. B-I. Dalenbäck. 'A New Model for Room Acoustic Prediction and Auralization'. Doctoral
Thesis. Chalmers University of Technology. (1995).
16. J-M. Jot. ‘Synthesizing Three-Dimensional Sound Scenes in Audio or Multimedia
Production and Interactive Human-Computer Interfaces.’ Proc. 5th Int. Conf. Interface to
Real & Virtual Worlds. (1996).
17. S. Pelzer et al. ‘Interactive real-time simulation and auralization for modifiable rooms’, Proc.
International Symposium on Room Acoustics. Toronto (2013).
18. B-I. Dalenbäck. ‘Whitepaper regarding diffraction.’ Report. (2012).
19. M. Vorländer. Auralization: Fundamentals of Acoustics, Modelling, Simulation, Algorithms
and Acoustic Virtual Reality. Berlin: Springer Verlag. (2008), p. 213.
20. F. Antonacci, et al. 'Inference of Room Geometry’.
21. J.B. Allen and D.A. Berkley. 'Image method for efficiently simulating small-room acoustics',
J. Acoust. Soc. Am. 65, 943-950. (1979).
22. J. Borish. 'Extension of the image model to arbitrary polyhedra', Journal of the Acoustical
Society of America 75(6), 1827-1836. (1984).
23. J. H. Rindel. ‘Attenuation of Sound Reflections due to Diffraction’, Proc. Nordic Acoustical
Meeting. Aalborg (1986).
24. J.S. Bradley and G.A. Soulodre. 'The influence of late-arriving energy on spatial
impression', J. Acoust. Soc. Am. 97, 2263-2271. (1995).
25. T. Lokki. 'Throw away that standard and listen: your two ears work better', Proc.
International Symposium on Room Acoustics. Toronto (2013).
... Performance space designs were progressed over 2 years and included visits to existing spaces, auralizations in the ArupSoundLab and the development of design prototypes, as documented by the authors in (2,3,4). The work led to several venue spaces and a palette of contrasting acoustical experiences, which included: ...
Conference Paper
Full-text available
Western classical music and arts are traditionally experienced in the concert halls and museums of cosmopolitan areas, or during festivals. New trends are emerging however, with diverse audiences being drawn to installations or concerts taking place in unusual or alternative environments including found spaces, "pop-up" venues, or curated "destinations". In this vein, the Tippet Rise Art Center, located in the foothills of Montana's Beartooth Mountains, and recognized today as a new destination for art and music in America, juxtaposes music, architecture and large-scale sculptures with dramatic natural landscapes. The paper describes the planning and design process of the art center and how it was conceived, at the outset, from sound principles, to create a diverse palette of indoor and outdoor environments for art and music, including a new music barn, an outdoor concert shell, large-scale outdoor concrete sculptures and a natural amphitheater. The paper presents acoustical measurements conducted in the finished projects. The authors contrast the results with traditional acoustic metrics, highlighting the relevance of context and multisensory performance setups on audience engagement, and proposes, as conclusions, new models for considerations in the planning of new arts and culture facilities.
... Those targets are translated into geometrical constraints while shaping the concert hall. The method generates any allowed shape that will achieve the acoustic targets, which can vary for example from a maximal ITDG ( " Initial Time Delay Gap " , see for example Beranek [7]) for side reflectors to the target acoustic signature of the room, as explored by Bassuet and Boning [8]. Sometimes real-time and interactive iterative design is quintessential for the design process, for example during workshops with architects. ...
Full-text available
The architectural design of many recent opera houses and concert halls has moved on from the traditional combination of parterre and balconies, to increasingly complex 3D concepts in which the traditional “acoustic” elements (balcony fronts, soffits, ceilings...) are no longer easily recognisable as individual elements. Consequently, the line between acoustics and architecture is becoming even thinner and more diffuse. This dynamism leads to the problem that by the time that the acoustic “checking” (by means of computer and/or scale models) of the design is finished, the room shape has already been determined to a large degree, driven by design factors other than acoustics. To face this challenge, we believe it is necessary to blend acoustic analysis in real time (or near real time) with architectural design tools. In addition, this establishes a collaborative forum in which both the architect and acoustic consultant can push each other’s creativity boundaries forward. Architectural parameterisation tools, such as Grasshopper, facilitate design in which both architecture and acoustic analysis and optimization are advanced simultaneously. Recent examples of projects with real-time 3D optimizations are presented.
Full-text available
An important factor in our appreciation of music in a hall is the perception of the spatial distribution of sound, influenced by room shape and form. This paper investigates new techniques for visualizing 3D impulse responses and two new spatial indicators are proposed: LH (ratio of low lateral versus high lateral energy), FR (ratio of front lateral versus rear lateral energy). Different room shape characteristics are illustrated from B-format measurements conducted in a selection of famous music spaces such as old and new recital and concert halls, and sacred music spaces.
Conference Paper
Impulse response of a room carries abstract information on the size and shape of that room. Acoustics research has focused on obtaining reflectograms from these physical characteristics using acoustical modelling techniques. However, estimation of the physical structure is also possible by analyzing the impulse responses for early reflections, echo amplitudes, arrival times and reverberation time. This paper proposes a new method to estimate the physical structure of a room using the features of directional impulse response (DIR) measurements obtained by applying the MLS-technique with a directional loudspeaker and a soundfield microphone. The results for 4 rooms were also displayed.
Acoustic scene reconstruction is a process that aims to infer characteristics of the environment from acoustic measurements. We investigate the problem of locating planar reflectors in rooms, such as walls and furniture, from signals obtained using distributed microphones. Specifically, localization of multiple two- dimensional (2-D) reflectors is achieved by estimation of the time of arrival (TOA) of reflected signals by analysis of acoustic impulse responses (AIRs). The estimated TOAs are converted into elliptical constraints about the location of the line reflector, which is then localized by combining multiple constraints. When multiple walls are present in the acoustic scene, an ambiguity problem arises, which we show can be addressed using the Hough transform. Additionally, the Hough transform significantly improves the robustness of the estimation for noisy measurements. The proposed approach is evaluated using simulated rooms under a variety of different controlled conditions where the floor and ceiling are perfectly absorbing. Results using AIRs measured in a real environment are also given. Additionally, results showing the robustness to additive noise in the TOA information are presented, with particular reference to the improvement achieved through the use of the Hough transform.
The book provides an up-to-date description of the methods used for fitting experimental data, or to estimate model parameters, and to unify these methods into the Inverse Problem Theory. The first part of the book deals with problems and describes Maximum likelihood, Monte Carlo, Least squares, and Least absolute values methods. The second part deals with inverse problems involving functions. Theoretical concepts are emphasized, and the author has all the useful formulas listed, with many special cases included. The book serves as a reference manual.
Imagine that you are blindfolded inside an unknown room. You snap your fingers and listen to the room's response. Can you hear the shape of the room? Some people can do it naturally, but can we design computer algorithms that hear rooms? We show how to compute the shape of a convex polyhedral room from its response to a known sound, recorded by a few microphones. Geometric relationships between the arrival times of echoes enable us to "blindfoldedly" estimate the room geometry. This is achieved by exploiting the properties of Euclidean distance matrices. Furthermore, we show that under mild conditions, first-order echoes provide a unique description of convex polyhedral rooms. Our algorithm starts from the recorded impulse responses and proceeds by learning the correct assignment of echoes to walls. In contrast to earlier methods, the proposed algorithm reconstructs the full 3D geometry of the room from a single sound emission, and with an arbitrary geometry of the microphone array. As long as the microphones can hear the echoes, we can position them as we want. Besides answering a basic question about the inverse problem of room acoustics, our results find applications in areas such as architectural acoustics, indoor localization, virtual reality, and audio forensics.