Artificial Neural Network Algorithm Based on Speech Recognition System To Improve The Utterance Rate Of Speech In Neural Network

Abstract

The investigation to which recognition systems can train the model parameters to produce the best class of discrimination will primarily determine their ability to correctly identify speakers based on their speech waveform distribution. This report details the outcomes of an effort to identify each speaker's voice using their distributed continuous voice waveform employing the coupled artificial neural network frame works. For discriminative classification and training, a feed-forward multi-layer ANN structure with 30 hidden neurons was used. This model created scores that were moved to best match the speech features. The decision system uses coefficient of correlation analysis to assess how well speech features match known speakers. frames from the ANN structures that describe the detected speaker. Investigations were performed out using spoken utterances from 30 distinct speakers to verify the system's performance (7 males and 3 females).For cases of trained voice utterances, system performance demonstrated average recognition rates of 95 percent for 1-word utterances and 5 percent when the length of the utterances was raised to 1 words. For 1-word utterances with unknown speakers, a recognition rate of 98% was reached.