Acoustic Simulation in Dynamic Environments for Robot Audition

This paper addresses acoustic simulation in dynamic environments for robot audition. For such environments, we consider three cases, that is, a moving microphone, a moving sound source, and a combination of the two. The proposed method simulates a dynamic environment by assuming that a motion trajectory of a microphone and/or a speaker can be discretized. We validated the proposed method through the accuracy of the simulated signals in terms of frequency and volume, and the performance of automatic speech recognition (ASR) with an acoustic model trained by simulated speech signals. The experimental results showed that the proposed method can simulate the sound properties of volume and frequency in dynamic environments well. The performance of ASR is improved with the acoustic model trained with the simulated speech signals.