Performance of PSO and GWO Algorithms Applied in Text-Independent Speaker Identification

In this paper, we analyze the performance of two bio-inspired algorithms applied in text-independent speaker recognition through voice signal. The analyzed algorithms are particle swarm optimization and grey wolf optimization. The complete methodology described in this paper was specifically developed in the context of this work. First, a widely known model of the speaker is determined based on discrete transfer functions. Then a method of estimation of the input signal is determined. The bio-inspired algorithms are custom-developed and applied to parameterize the transfer functions based on the models. The proposed method is composed by three major parts, first the fitness used in the bio-inspired algorithms is created based on the cross-correlation. Second, a method to create a database with speakers’ identities is proposed, and third, a method to compare the characteristics of the speaker is proposed, to identify or distinguish two different speakers. Finally, experiments were made considering 4 speakers with 2 speech each, a representation of the identity of each speaker was created through both algorithms, totalizing 16 entries on the database. The experiment had a total of 240 runs, comparing the entries to each other. Results show that all comparison results were accurate. The algorithms identify the speaker even when two different speeches were compared, and, as expected, distinguish when two different speakers were compared.