Endpoint Detector Algorithm for Speech Recognition Application

Abstract

This paper proposes a Speech Recognition Application (SRA) for speech recognition with an accuracy location of endpoints on a speech signal-recording, with an Adaptive Equalization (AE), Energy Pulse Detector (EPD), Feature Extraction (FE) and a decision rule Database MySQL. In order to localize an endpoint in the signal, a signal recording has been processed by the Endpoint Detector Algorithm (EDA) to localize the speech segments and non-speech segments by slicing the noise and the zero rates of signal, this technique could also found an isolated speech on the recording-signal by detecting which word is considered from the same speech or a different speech. The decision rule Database MySQL compares the input signal with all the stored signals saved in the database and identified if that speech is already on the system. To evaluate the performance of the SRA system, speech recognition experiments are carried in computer simulation at non-real environments by a single microphone. This result shows that the proposed endpoint algorithm attains a higher level recognition on speech signals in the SRA system.