Speech Recognition Using ARMA Model and Levenberg-Marquardt Algorithm
Files
TR Number
Date
2024-08-01
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Springer
Abstract
Autoregressive Moving Average (ARMA) is a simple linear model with memory that can be used for speech recognition problems. This is why, this paper utilized the derivation of ARMA model for the speech recognition. The flexibility of ARMA model helps in derivation of an accurate model that recognizes the pronunciation of letter B. The Generalized Partial Autocorrelation (GPAC) analysis has been used for the preliminary identification and the Maximum Likelihood Estimator (Levenberg-Marquardt) is used for the parameter estimations. Several models have been developed to recognize the letter B that are pronounced by a lady 30 times. The simplest model has been chosen at the end. The accuracy of the final model has been checked using χ2 test.
Description
Keywords
Speech recognition, Nonlinear optimization, Residual analysis