Soft Computing Techniques for Classification of Voiced/Unvoiced Phonemes

摘要

A method that uses fuzzy logic to classify two simple speech features for the automatic classification of voiced and unvoiced phonemes is proposed. In addition, two variants, in which soft computing techniques are used to enhance the performance of fuzzy logic by tuning the parameters of the membership functions, are also presented. The three methods, manually constructed fuzzy logic (VUFL), fuzzy logic optimized with genetic algorithm (VUFL-GA), and fuzzy logic with optimized particle swarm optimization (VUFL-PSO), are implemented and then evaluated using the TIMIT speech corpus. Performance is evaluated using the TIMIT database in both clean and noisy environments. Four different noise types from the AURORA database-babble, white, restaurant, and car noise-at six different signal-to-noise ratios (SNRs) are used. In all cases, the optimized fuzzy logic methods (VUFL-GA and VUFL-PSO) outperformed manual fuzzy logic (VUFL). The proposed method and variants are suitable for applications featuring the presence of highly noisy environments. In addition, classification accuracy by gender is also studied.