On Four Metaheuristic Applications to Speech Enhancement—Implementing Optimization Algorithms with MATLAB R2018a

The 2018 Conference on Computational Linguistics and Speech Processing ROCLING 2018, pp. 266-275 ©The Association for Computational Linguistics and Chinese Language Processing On Four Metaheuristic Applications to Speech Enhancement －Implementing Optimization Algorithms with MATLAB R2018a Su-Mei Shiue Department of Computer Science and Information Engineering National Taipei University of Technology t102599005@ntut.edu.tw Lang-Jyi Huang Department of Computer Science and Information Engineering National Taipei University of Technology t8599003@ntut.edu.tw Wei-Ho Tsai Department of Electronic Engineering National Taipei University of Technology whtsai@ntut.edu.tw Yen-Lin Chen Department of Computer Science and Information Engineering National Taipei University of Technology ylchen@csie.ntut.edu.tw Abstract This study aims to firstly implement four well-known metaheuristic optimization algorithms which, among other things, are utilized on adaptive filter design, for dual-channel speech enhancement, with voice communication devices. The implementations are conducted in a simulation fashion using MATLAB code under its newly release version of R2018a. Lately, the study takes a closer look at these four optimization methods, based on learning from the literature of the research society, on their pros and cons while applying to speech enhancement. These four methods are, namely, (1) Accelerated Particle Swarm Optimization (APSO), (2) Gravitational Search Algorithm (GSA), (3) a hybrid algorithm of PSO (Particle Swarm Optimization) and GSA (called hybrid PSOGSA), and (4) Bat Algorithm (BA). This 266 study performs the said implementations successfully and obtains useful experimental results which contributes to building a solid foundation for more advanced research in the near future. Besides, the implementations made by the study confirm the correctness of many a previous research works which claim that these four algorithms show better performance on improved speech quality and intelligibility as compared to that of the existing standard PSO (SPSO) based speech enhancement approach. Keywords: Speech Enhancement, APSO, GSA, PSOGSA, BA. 1. Introduction The concise characteristics and advantages of metaheuristic optimization methods as discussed in this paper for speech enhancement are listed in Table 1 below. Table 1. Important metaheuristic optimization methods for speech enhancement Name Characteristics PSO PSO does not require that the problem of optimization to be differentiated as required by classical optimization methods such as gradient descent and quasi Newton methods. Therefore, PSO can be used in optimization problems that are partly irregular, variable over time, etc. [5]. Simple conception, quick velocit significant SNR ratio for higher acc person will percept/ recognize the w GSA An algorithm inspired by Newton‟s laws of motion and gravitation force [1]. Updating is done by considering th performance than PSO [3]. PSOGSA A hybrid algorithm of PSO and GSA [3]. Combines the advantages of GSA a designing the adaptive filter in nois APSO Developed by Xin she Yang in 2010. APSO is simpler to implement and has faster convergence when compared to the standard PSO (SPSO) algorithm. [2] APSO based speech enhancement i speech quality and intelligibility me BA (1) Frequency tuning: BA uses echolocation and frequency tuning to solve problems. (2) Automatic zooming: BA has a distinct advantage over other metaheuristic algorithms. of automatically zooming into a region where promising solutions have been found …

Content truncated to 500 words