Speech recognition using Taylor-gradient Descent political optimization based Deep residual network. (March 2023)

• Record Type: Journal Article
• Authors: V․H․, Arul; Marimuthu, Ramalatha
• Is Part Of: Computer speech & language. Volume 78(2023)
• Publication Date: 2023-03
• Subjects: Speech recognition -- Bark frequency cepstral coefficients -- Deep residual network -- Tonal power ratio -- Spectral centroid; Speech processing systems -- Periodicals; Automatic speech recognition -- Periodicals; Computers -- Periodicals; Linguistics -- Periodicals; Speech-Language Pathology -- Periodicals; Traitement automatique de la parole -- Périodiques; Reconnaissance automatique de la parole -- Périodiques; Automatic speech recognition; Speech processing systems; Electronic journals; Periodicals; 006.454
• DOI: 10.1016/j.csl.2022.101442
• Languages: English
• ISSNs: 0885-2308

Abstract

Highlights: Developed Taylor GDPO-based DRN for speech recognition : The Taylor GDPO-based DRN is devised for recognizing the speech using audio signals. Here, the training of DRN is performed with proposed Taylor GDPO, which is devised by combining Taylor series, GD, and PO. The proposed Taylor GDPO-based DRN offered effective performance with highest accuracy of 96.93%, smallest FAR of 2.438%, smallest FRR of 2.101%, smallest MSE of 0.038. Abstract: A Speech is an effective and most favoured communication mode amongst the human. It is normal for people to anticipate speech interfaces with computer. For real-time intelligent applications, it is necessary that the machine can hear, understand, investigate, and take action upon receiving the input information from speaker. This can be taken by introducing an Automatic Speech Recognition system, which translates an audio signal into a written text or a command without understanding what has been recognized. Several methods are designed for speech recognition, but accuracy is a most challenging task. Hence, this paper develops a novel Taylor Gradient Descent Political Optimizer (Taylor GDPO) based deep learning model for speech recognition. A developed Taylor GDPO is obtained by integrating Taylor series, Gradient Descent (GD) and Political Optimizer (PO). Firstly, pre-processing of an input signal is done and the features are extracted. Then, the extracted features are given as input to the Deep Residual Network (DRN), which is trained by the developed Taylor GDPO. A developed model provided effectual efficiency with the highest accuracy of 96.93%, smallest False Acceptance Rate (FAR) of 2.44%, smallest False Rejection Rate (FRR) of 2.10%, smallest Mean squared error (MSE) of 0.038.