Microphone arrays have been widely used for hands-free speech acquisition systems such as teleconferencing or automatic speech recognition. They typically require sound source localization for subsequent multichannel signal processing algorithms such as beamforming for enhancing speech signals. Steered response power with phase transform (SRP-PHAT) source localization has been the most popular method thanks to its robustness against reverberation. In most cases, the SRP-PHAT requires exhaustive search methods, which makes its real-time implementation a challenging task. In order to address this
problem, this paper presents a GPU (Graphic Processor Unit) implementation that takes advantage of several parallel steps involved the the computation of the SRP-PHAT.