Improvement a Transcription Generated by an Automatic Speech Recognition System for Arabic Using a Collocation Extraction Approach

Abstract

. The following study propose a novel heuristic to improve an automatic speech recognition system for Arabic language. Our heuristic relies on the col-laboration of two approach: the first one ensures the extraction of collocations from a voluminous corpus then stores them in a database. It uses a combination of several classical measures to cover all aspects of a given corpus in order to exclude bigrams having a high probability of occurring together. The second one constructs a search space on the relations of semantic dependence of the output of a recognition system then, it applies phonetic filter so as to select the most probable hypothesis. To achieve this objective, different techniques are deployed, such as the word2vec or the language model RNNLM in addition to a phonetic pruning system. The obtained results showed that the proposed approach allowed improving the precision of the system.