Endpoint detection of speech signal using neural network
Aini Hussain, Salina Abdul Samad, Liew Ban Fah · 2002
This paper highlights the artificial neural network (ANN) approach to perform the endpoint detection process, which involves the segmentation of speech signals from non-speech signals. Two ANN models have been proposed to perform endpoint detections of isolated digit utterances spoken in the Malay language: multilayer perceptron (MLP) and adaptive linear network (ADALINE). Results obtained from the ANN models are acoustically verified, visually checked and compared to the conventional method of endpoint detection. It was found that the endpoint detection accuracy using the MLP approach is very high and encouraging.