AUD-SWIPE-P: A Parallelization of the AUD-SWIPE Pitch Estimation Algorithm Using Multiple Processes and Threads

In this work we implement a parallel version of the pitch estimation algorithm Aud-SWIPE. The new implementa-tion, Aud-SWIPE-P produces a remarkable acceleration, making it usable in applications where it was not efficient before because of its high computational cost. We paral-lelized the algorithm using multiple threads and processes to speed up different steps of the algorithm, and take advan-tage of different architectures. The goal for Aud-SWIPE-P’s performance was to achieve an execution time below the signal duration, opening the possibility to create a real-time software using Aud-SWIPE as pitch estimation algo-rithm. The implementation was evaluated running the al-gorithm in different real-world scenarios. The acceleration achievedwas around 4.4, comparedwith the sequential ver-sion of the algorithm. In sound files with a sampling fre-quency of 10 kHz or less, the real time execution goal was fully achieved.