Towards an Hybrid Approach for Semantic Arabic Spontaneous Speech Analysis

The automatic speech understanding aims to extract the useful meaning of the oral utterances. In this paper, we propose a hybrid original method for a robust automatic Arabic speech understanding. The proposed method combines two approaches usually used separately and not considered as complementary. This hybridization has the advantage of being robust while coping with irregularities of oral language such as the non-fixed order of words, selfcorrections, repetitions, false departures which are called disfluencies. Through such a combination, we can also overcome structuring sentence complexities in Arabic language itself like the use of conditional, concession, emphatic, negation and elliptical forms. We provide, in this work a detailed description of our approach as well as results compared with several systems using different approaches separately. The observed error rates suggest that our combined approach can stand a comparison with concept spotters on larger application domains. We also present, our corpus, inspired from MEDIA and LUNA project corpora, collected with the Wizard of Oz method. This corpus deals with the touristic Arabic information and hotel reservation. The evaluation results of our hybrid spontaneous speech analysis method are very encouraging. Indeed, the obtained rate of F-Measure is 79.98%.