Embedding and Beamforming Network for Sound Source Localization in Spherical Harmonic Domain

Sound source localization (SSL) is an essential task in many applications involving speech separation and enhancement. As such, speaker localization with microphone arrays has received significant research attention. While traditional signal processing (SP)-based SSL methods provide analytic solutions under specific signal and noise assumptions, recent deep learning (DL)-based methods have significantly outperformed them. However, their success depends on extensive training data and substantial computational resources. Moreover, they often rely on large-scale annotated spatial data and may struggle to adapt to reverberation and noise. To mitigate these challenges, we propose a novel embedding and beamforming network (EaBNet)-based SSL approach, termed SSL-EaBNet, which computes the spatial covariance matrix (SCM) of the observed signal after masking, compares the error with the target SCM, and performs error backpropagation to update the model weights. In particular, EaBNet has high computational efficiency and real-time inference capabilities, and its offline training process meets the requirement for real-time application. Empirical results in the public localization and tracking (LOCATA) dataset demonstrate the superior performance of our proposal, achieving a localization accuracy (ACC) of 91.49%, surpassing other competitive methods.