Toward an automatic speech recognition system for amazigh-tarifit language

Abstract

This work aims at contributing to the Amazigh language Automatic Speech Recognition (ASR). We have studied and realized an automatic speech recognition system, using an environment totally based on the Amazigh-Tarifit language. In this framework, we have first constructed a new Amazigh-Tarifit speech corpus, which was used to assess and make the results of this work undergo a test. In fact, this paper has two objectives: The first one is to collect a Medium-Vocabulary Isolated Word speech corpus, which will serve as a corpus for the Amazigh speech researchers. The second target is to develop an Amazigh ASR system using this speech corpus (187 distinct isolated-words). The speech corpus was recorded by 50 individuals (25 males and 25 females) Amazigh-Tarifit native speakers. The system was evaluated on a speaker-independent approach. The tests were carried out basing essentially on two parameters: the Gaussian Mixture Distributions (GMM), and tied states (senones). The Word Error Rate (WER) achieved 8, 20%.