Parallel SRP-PHAT for GPUs. (January 2016)

• Record Type: Journal Article
• Authors: Lee, Taewoo; Chang, Sukmoon; Yook, Dongsuk
• Is Part Of: Computer speech & language. Volume 35(2016)
• Page Start: 1
• Page End: 13
• Publication Date: 2016-01
• Subjects: Sound source localization -- SRP-PHAT -- GPUs; Speech processing systems -- Periodicals; Automatic speech recognition -- Periodicals; Computers -- Periodicals; Linguistics -- Periodicals; Speech-Language Pathology -- Periodicals; Traitement automatique de la parole -- Périodiques; Reconnaissance automatique de la parole -- Périodiques; Automatic speech recognition; Speech processing systems; Electronic journals; Periodicals; 006.454
• DOI: 10.1016/j.csl.2015.05.002
• Languages: English
• ISSNs: 0885-2308

Abstract

Highlights: Parallelization of SRP-PHAT algorithms. Optimal implementation of the proposed parallel SRP-PHAT for GPGPUs. Achievement of 1.5–6 times speedup in sound source localization. Abstract: The steered response power phase transform (SRP-PHAT) is one of the widely used algorithms for sound source localization. Since it must examine a large number of candidate sound source locations, conventional SRP-PHAT approaches may not be used in real time. To overcome this problem, an effort was made previously to parallelize the SRP-PHAT on graphics processing units (GPUs). However, the full capacities of the GPU were not exploited since on-chip memory usage was not addressed. In this paper, we propose GPU-based parallel algorithms of the SRP-PHAT both in the frequency domain and time domain. The proposed methods optimize the memory access patterns of the SRP-PHAT and efficiently use the on-chip memory. As a result, the proposed methods demonstrate a speedup of 1276 times in the frequency domain and 80 times in the time domain compared to CPU-based algorithms, and 1.5 times in the frequency domain and 6 times in the time domain compared to conventional GPU-based methods.