Room transfer function estimation with variable boundary conditions using artificial neural networks

Abstract

One of the sound field control techniques, the sound focusing is the concept of sound field control to make acoustic bright zone and acoustic dark zone. In the acoustic bright zone, users could hear well the sound from the speaker array. And users could not hear well the sound from the speaker array in the acoustic dark zone. To make theses acoustic dark and bright zone, one need to get room transfer functions between each speaker in the speaker array and target points. Room transfer functions are given by the environment. So Getting a room transfer function is important to conduct sound filed control. To obtain a room transfer function, there were various measurement methodologies suggested. But there were immediacy problems for the measurement methodologies. Because the reliability and the measurement cost have the trade-off relationship. For example, if one conducts small number of repetitions of measurement, then one would get a relatively not reliable result while spending relatively cheap measurement cost. Otherwise, if one conducts measurement with the large number of repetitions, relatively reliable result could be obtained while taking expensive measurement cost. Thus, if one wants to get a room transfer function immediately by using measurement, reliability would not be guaranteed. So in this study, the methodology of estimation of the room transfer function under the variable boundary conditions using artificial neural networks was suggested instead of measurement. Measurement technique was used for collecting training data from the real environment. The methodology of estimation of room transfer functions by using artificial neural networks was conducted and compared with interpolation method. The performance of the artificial neural networks was shown that it is better than the interpolation method especially for phase estimation. At the same time, one can obtain immediacy to get room transfer functions by using estimation of acoustic room transfer function instead of measurement and acoustic simulation techniques.