Open AccessPitch tracking algorithm based on evolutionary computing with regularisation in very low SNR
Author(s) -
Zhang Xiaoheng,
Li Yongming
Publication year - 2018
Publication title -
the journal of engineering
Language(s) - English
Resource type - Journals
ISSN - 2051-3305
DOI - 10.1049/joe.2018.8290
Subject(s) - evolutionary algorithm , computer science , algorithm , fitness function , regularization (linguistics) , particle swarm optimization , tracking (education) , genetic algorithm , mathematical optimization , mathematics , artificial intelligence , machine learning , psychology , pedagogy
The authors present PTEAR_VLSNR (Pitch Tracking basing on Evolutionary Algorithm with Regularization at Very Low SNR), a pitch tracking algorithm for speech in strong noise. The algorithm builds a pitch enhancement and extraction model, which enhance the pitch by a matched filter, and to further deal with strong noise, the optimal factor was proposed, which can be optimised globally by the evolutionary computing. Specially, regularisation constraint of fitness function was applied to enhance the generalisation ability. Temporal dynamics constraints are used to improve the tracking rate and the voicing decision can be optimal by evolutionary computing similarly. In addition, the balance of optimisation accuracy and time cost were considered. In experiments, genetic algorithm and particle swarm optimisation with two‐norm term were represented as evolutionary algorithms with regularisation. At last, they compare the performance of the algorithm and other representative algorithms. The experimental results show that this proposed algorithm performs well in both high and low signal‐to‐noise ratios (SNRs).