AUD-SWIPE-P: A PARALLELIZATION OF THE AUD-SWIPE PITCH ESTIMATION ALGORITHM USING MULTIPLE PROCESSES AND THREADS

摘要

In this work we implement a parallel version of the pitch estimation algorithm Aud-SWIPE. The new implementation, 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 parallelized the algorithm using multiple threads and processes to speed up different steps of the algorithm, and take advantage of different architectures. The goal for Aud-SWIPEP's performance was to achieve an execution time below the signal duration, opening the possibility to create a realtime software using Aud-SWIPE as pitch estimation algorithm. The implementation was evaluated running the algorithm in different real-world scenarios. The acceleration achieved was around 4.4, compared with the sequential version of the algorithm. In sound files with a sampling frequency of 10 kHz or less, the real time execution goal was fully achieved.