Analysis of Feature Extraction Techniques for Myanmar Automatic Speech Recognition

Automatic Speech Recognition (ASR)system is to accurately and efficiently convertspeech signal into a text message independent ofdevice, speaker or the environment. Featureextraction is the second component of automaticspeech recognition systems which extract theinformation from the speech frame. The featureextraction is needed because the raw speechsignal contains information besides the Linguisticmessage and has a high dimensionality. Theprimary objective of feature extraction is to findrobust and discriminative features in the acousticdata. The recognition module uses the speechfeatures and the acoustic models to decode thespeech input and produces text results with highaccuracy. There are several techniques for featureextraction , this paper is the comparative analysisof four feature extraction techniques of FilterBank (FBank), Mel Frequency CepstralCoefficient (MFCC), Perceptual Linear Predictive(PLP) and Gammatone Frequency CepstralCoeffcieint (GFCC) for Myanmar continuous ASR.The experimental result shows that with theclassification method Gaussian Mixture Model(GMM). The better performance of featureextraction method is to support for Myanmar ASR