Dynamic analysis for isolated Arabic words recognition
Z. Hachkar1, B. Mounir1, A. Farchi2*, J. Abbadi3
1 EST Safi, route dar si aissa, Safi, Moroco.
2 FST Settat, Km 3 B.P :577, 26000 Settat, Moroco.
3 EMI Rabat, Ibnsina Avenue, B.P. 765Agdal Rabat, Moroco.
* Corresponding author: E-mail: firstname.lastname@example.org
Received: 17 June 2010; revised version accepted: 15 March 2011
The aim of this paper is to study the feasibility of using Dynamic Time Warping algorithm to implement a system of automatic recognition of Arabic isolated words. The extraction technique used to characterize the signal is "Mel Frequency cepstral coefficients or MFCC". Other prosodic parameters, such as energy, first and second drifts of MFCC, were concatenated to basic MFCC coefficients in order to search for more effective signal characteristics.
A corpus of apprenticeship and of test pronounced by several Moroccan speakers is established in order to evaluate system performances. The better recognition rate (86%) was obtained while using, as characteristics, 12 coefficients MFCC, an energy coefficient, and 26 coefficients obtained from the first and second drifts of the MFCC coefficients and energy.
Keywords: Speech recognition; Log Energy; DTW; MFCC.