Continuous formal Arabic speech recognition system based on hidden Markov model
Sa’ed Abed, Hanem Ellethy, Mohammad H. Alshayeji · 2017
Speech recognition is a mechanism to recognize words and phrases in any language and translate them to a machine-readable layout. Recently, Automatic Speech Recognition (ASR) systems are used widely in many applications like translating speech to text, home security systems, and military applications. Unfortunately, the speech recognition field for the Arabic language is not yet mature and it needs more development. The main objective of this paper is to implement and evaluate a continuous speech recognition system for the formal Arabic language using MelFrequency Cepstral Coefficients (MFCCs) and Hidden Markov Model (HMM) with Gaussian Mixture Model (GMM). Adding the energy feature to the MFCCs twelve features for each frame has a good impact on the accuracy for the continuous speech. In addition, using GMM was effective as a post processing step before calculating the HMM parameters for training. A recognition rate is enhanced to be 94% as a result of tuning parameters and using different techniques. The Arabic speech recognition system is built using MATLAB based on HMM toolbox. A comparison between our work and the state of the art works is presented to show the acceptability and intelligence (accuracy) of our work.