Automatic Speech Recognition on Spontaneous Interview Speech

Abstract

This paper presents a spontaneous speech recognition system for Myanmar language. Automatic speech recognition (ASR) on some controlled speech has achieved almost human performance. However, the performance of spontaneous speech is drastically decreased due to the diversity of speaking styles, speak rate, presence of additive and non-linear distortion, accents and weakened articulation. In this study, we built a recognizer for Myanmar Interview speech by using the classical Gaussian Mixture Model based Hidden Markov Model (HMM-GMM) approach. We invested that the effect of variation on acoustic feature and number of senones and Gaussian densities on Myanmar Interview speech. According to these experiments, we achieved the best Word Error Rate (WER) of 20.47%.