Improvement of the Direction-of-Arrival Estimation Method Using a Single Channel Microphone by Correcting a Spectral Slope of Speech
Masaki Ikeuchi, Hiroki Tanji, Takahiro Murakami · 2022 Asia-Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC) · 2022
To estimate the direction-of-arrival (DOA) of an acoustic signal with a single channel microphone, a method for estimating a notch in a spectrum that is generated according to the DOA was proposed. The system used in this method is composed of a single channel microphone and a flat wall installed near the microphone. The notch in the spectrum according to the DOA is generated by interference between a direct wave and a reflected wave from the wall. In the conventional method, assuming that a source signal follows a normal distribution, the notch frequency according to the DOA is estimated by maximum likelihood estimation. In order to extend this method to general acoustic signals such as speech, the method of averaging spectra of an observed signal over time was also proposed. However, relatively large estimation errors remain because an undesirable spectral slope appears in the time-averaged spectrum. Thus, in this paper, we propose a method to correct such a spectral slope. Our approach is based on the fact that the time-averaged spectrum tends to resemble a power-law distribution. Simulation results show that correcting the spectral slope improves the performance of DOA estimation.