A New Sound Source Localization Approach Using Stereo Directional Microphones
Donggeon Lee, Beomhee Jang, Sungbin Im, Jinzi Song · 2019
This paper proposes a new Sound Source Localization (SSL) algorithm using the features of 2-channel directional microphones. Noise reduction techniques and voice activity detection (VAD) are used to calculate the root mean square (RMS) values of the signal received on each channel. In the noise reduction stage, band pass filter, improved minima controlled recursive averaging (IMCRA), and multichannel Wiener filter (MWF) are applied to reduce the ego noise from UAV. After data preprocessing and RMS value computation, specific grouping phenomena is observed according to the angle. In this study, we employ K-medoids, one of the clustering techniques, to classify the groups. The data used in this study to investigate the proposed approach are collected as follows: The angle between the speaker and the microphone varies from 0 to 180 degrees with a 60degree increment, and the distance does from 5 to 20 meters. The distance between the microphone and drone is set to 0.3 through 0.8 meters. 50 data are randomly selected for the evaluation. Since the proposed approach is based on the RMS ratios in the speech presence sections, it can be applied to the situation in which the time delay of arrival (TDOA) methods are difficult to be applied.