Isolated Myanmar Speech Recognition via ANN

Abstract

Automatic Speech Recognition (ASR) is a popular and challenging area of research in human computer interaction. This paper presents an isolated Myanmar speech recognition system that is speaker dependent as well as speaker independent and developed by using Artificial Neural Network techniques. In this system, the Mel Frequency Cepstral Coefficients extracted from the manually preprocessed words are considered as the features to acoustically identify the speeches. Those features are then used to train and test the Backpropagation neural network model. This system uses a database of 2800 utterances (names of the cities in Myanmar) by 10 talkers (4 males and 6 females), from which 2400 utterances are used for training and 400 are used for testing and recognition. As per the experimental results, the proposed system achieved the recognition rate of about 93.5% for known speakers (i.e., speaker dependent) and 76.5% for unknown speakers (i.e., speaker independent).