154th ASA Meeting, New Orleans, Louisiana

[ Lay Language Paper Index | Press Room ]

Inverse Sound Rendering: capturing the acoustic properties of realistic interiors into a virtual model

Gabriel Pablo Nava - pablo@iis.u-tokyo.ac.jp
Yoichi Sato
Sinichi Sakamoto

Institute of Industrial Science
The University of Tokyo
Komaba 4-6-1, Meguro-ku, Tokyo
153-8505 JAPAN

Yosuke Yasuda
Graduate School of Frontier Science
The University of Tokyo
Kashiwanoha 5-1-5, Kashiwa-shi, Chiba
277-8563 JAPAN

Popular version of paper 4aAA1
Presented Friday morning, November 30th, 2007
154th ASA Meeting, New Orleans, Lousiana

Let us suppose that we are inspecting an empty room which we plan to transform into an office, a home studio, or simply we just plan to move into that room in a near future. We would probably have a hard time trying to imagine in advance how the new room would look and sound like after the changes are done, and in general, what the proper changes are in order to minimize expenses. Fortunately, nowadays there are powerful computer algorithms that allow us to visualize and auralize the effects of the changes in a virtual model. Although most of these simulation tools are accurate enough to approximate the real visual and audio effects, they have in common the following problem: in order to achieve high quality simulations, the acoustic and lighting properties of the objects need to bee specified in the virtual model as they are in the real world. For this reason, researchers in computer graphics have recently developed a technique known as inverse global lighting rendering to measure the light reflectance properties on the surface of the objects in a room from a set of photographs. Furthermore, using the measured properties, they have been able to synthesize realistic images of a room which includes objects and/or changes that never existed in the real scenery. Unfortunately for audio, no computer graphic method had been developed before to measure the acoustic properties of the surfaces in interiors. That became the motivation of the research project introduced here.

In acoustics, the traditional principle to measure the acoustic properties of the materials in their original context consists of comparing the direct sound coming from a speaker with the sound reflected on the test surface at different positions (usually two, where the measuring microphones are placed; see Figure 1). This method, although widely used, has constraints that prevent its use in most practical applications: 1) the test surface should be large and flat, 2) the microphones should be placed at optimum positions near the surface, and 3) the measurements should be performed in a wide space. Hence, this technique becomes troublesome if inaccessible complex-shaped surfaces, such as those in a real room, are considered.

Fig. 1. Traditional method for measurement of the acoustic properties of the materials.

In order to overcome the problems mentioned above, we have proposed an approach to measure the acoustical properties of the interior surfaces of a room. This new approach takes as input 1) the 3D model of the room, 2) the characteristics of the sound source (a speaker), and 3) a set of sound samples measured at random locations in the interior field. The setup of the system is illustrated in Fig. 2.

Now let us suppose that we would like to estimate the ability to resist the passage of sound of the interior surfaces of the room depicted in Figure 2. The measurement procedure is as follows: 1) specify the sound frequency at which the acoustic impedance is desired, 2) emit a tone of the selected frequency through the speaker, 3) start the measurement process, 4) move the microphone freely along the room until the necessary number of measurements of sound pressure have been acquired by the system, 5) stop the measurement process and start the data processing. The computer processes the data and puts out an estimate of the acoustic impedances.

Fig. 2. Setup of the proposed method to measure the acoustic properties of the surfaces in a room.

In contrast to traditional methods, the proposed technique can be applied to the interior surfaces of an arbitrary-shape room. On the other hand, the measurement has to be repeated for each frequency of the audible spectrum, but in exchange, the acoustic properties of not only one but all the surfaces in the room (provided that their geometry is known). Moreover, in practice, the system integrates computer vision tools (such as 3D stereo tracking) to make it easier to acquire sound samples at random positions while moving a single microphone freely in the room.

In the development stage, the method has been tested with computer simulations with realistic geometries such the one shown in Fig. 3, showing successful results. Furthermore, before testing the method in a real room, experiments have been performed in a controlled environment with the purpose of confirming its applicability in practical situations. As can bee seen in Fig. 4, the experimental setup consists of a scaled model of a rectangular room (a 3 cm-thick acrylic box) in which different types of surfaces have been set.

Fig. 3. Office model used for the test simulations. (units in meters).

Fig. 4. Scaled-model room used for the preliminary experiments with the proposed method.

Although the system is in an experimental state, the results from the preliminary experiments suggest that the method may be successfully used in real interiors to recover the acoustic properties of the surfaces, provided that the property of the surface is fundamentally determined by the acoustic effect at the point where the sound waves hit the surface (a condition which is known as local reaction). This condition is frequently found in empty rooms, therefore, when simulations of interiors are desired, this method is particularly useful to prescribe the acoustic properties of the model. In addition, while the development of this project is in an ongoing stage, efforts are been made to overcome limitations of the current implementation that keep it from being applied to large rooms due to the high computational requirements.

[ Lay Language Paper Index | Press Room ]