Comparative study and improvement of acoustic vectors extractors: Multiple streams applied to the recognition of Arabic numerals

Abstract

This paper exposes at the beginning most of the problems encountered in the automatic speech recognition, namely the complexity of speech signal, and the diversity of process of signal processing. For this, a review of general concepts in signal processing is inspected; after it has been specifically addressed the extraction phase, in order to conduct a thorough review of existing extraction methods in this area. The next task is reserved to a list of existing extractor's combinations, and finally an extractor's recombination is developed, with the aim of reaching a full treatment with a considerable amount of extracted vectors acoustic, in order to develop the recognition rate of Arabic numerals.