Robust front-end based on MVA processing for Arabic speech recognition

This paper presents a noise robust technique for arabic automatic speech recognition engine. The technique is based on Cepstral Mean and Variance Normalization (CMVN) plus Auto Regressive Moving Average (ARMA) filtering which is called MVA. MVA used as a post-processing module to Mel Frequency Cepstral Coefficients (MFCC), Relative Spectral-Perceptual Linear Prediction (RASTA-PLP) and Power Normalized Cepstral Coefficients (PNCC) features to improve the recognition accuracy. While an isolated Arabic word engine was designed and developed using the Hidden Markov Model (HMM) to perform the recognition process at the back-end. Experimental results on the Arabic database demonstrate that our method provides substantial improvements in recognition accuracy for all features. The results also demonstrate that RASTA-PLP outperforms PNCC and MFCC features for word correction and word accuracy.