Analysis of Speaker Verification Method Based on Hidden Markov Models for Continuous Speech System
Mahdi Fadil Khaleeland · International Journal of Analytical Experimental and Finite Element Analysis (IJAEFEA) · 2023
This paper aim to the effectiveness of speaker verification using prompted text. The progress and enhancement of ASV applications have significant implications, particularly considering their advantages in comparison to alternative biometric methodologies., support vector machines (SVM), Hidden Markov models (HMM), the generalized method of moments (GMM), artificial neural networks (ANN), and combination models are only some of the statistical models used by modern speaker recognition systems. Using a dataset collected in Turkish. The goal of this work was to create a continuous speech system using Hidden Markov Models (HMM) on a single mixed monophonic level, independent of any surrounding environment. Subsequently, appropriate speech data is used in the construction of both the client and world models. The text-dependent speaker verification method employs sentence Hidden Markov Models (HMMs) that have been concatenated for the designated text in order to authenticate speakers. The normalized log-likelihood is calculated in the verification stage by comparing the log-likelihood of the client model, which is derived using the Viterbi method and the world model. It is by subtracting these two log-likelihood values that we arrive at the normalized log-likelihood. Finally, a method for evaluating verification results is shown.