Single Sound Source Localization Using the SRP-PHAT Method

This article introduces the SRP-PHAT (Steered Response Power with Phase Transform) method for single sound source localization. SRP-PHAT determines the sound source position by calculating time difference of arrival (TDOA) measurements between microphone signals. This popular sound source localization technique is widely applied in speech recognition, music information retrieval, and other audio processing domains. The core algorithm involves computing phase-transform weighted cross-correlations between microphone pairs and steering the response power across potential source locations through delay-and-sum beamforming.

For practical implementation, consider a scenario where we need to locate a sound source in a room environment. By deploying multiple microphones at known positions, we can capture audio signals and calculate generalized cross-correlation with phase transform (GCC-PHAT) between each microphone pair. The implementation typically involves: 1) Preprocessing audio signals using frame-based processing with overlap, 2) Applying PHAT weighting in the frequency domain to emphasize phase information, 3) Computing steered response power for each candidate location on a search grid, 4) Identifying the maximum power position as the source location. This method enables accurate sound source positioning, facilitating better environmental awareness and event analysis in applications like smart homes and conference systems.