Abstract:Sound Source Localization (SSL) is a position-determining process of the sound source, which is based on necessary information such as the phase of the audio signal. Although an increasing number of studies have been proposed and used to solve the SSL problem, little attention has been paid to improving the performance of SSL under heavy echo conditions. When the significant echoes are superimposed on audio signals, the original phase of the sound source will be disrupted, which will significantly increase the difficulty of SSL. To minimize the effect of echoes on localization, a novel algorithm is proposed in this article. Firstly, a linear Acoustic Echo Cancellation (AEC) algorithm is used to eliminate a certain amount of echo. Then, a Pseudo Residual Acoustic Echo Suppression (PRAES) model is used to mask spectrogram and compensate for the speech phase. Finally, the Direction of Arrival (DOA) model is used for localization. The localization results integrating spectral masking and phase compensation are superior to that of other comparative experiments. Field tests illustrate that the proposed algorithm has better performance than that of the existing state-of-the-art algorithms.

Echo−influenced Source Localization Algorithm Using Frequency Masking and Phase Compensation