Wolof Speech Recognition Model of Digits and Limited-Vocabulary Based on HMM and ToolKit

Abstract

This paper is concerned with Automatic Speech Recognition (ASR) using trainable systems. The aim of this work is to build acoustic models for spoken language Wolof. This is done by employing Hidden Markov Models (HMM) and using the different lexicons and knowledge bases of Wolof to train their parameters. Acoustic modeling has been worked out at a phonetic level, allowing general speech recognition applications, even though a simplified task (natural number recognition and Limited-Vocabulary Speech) has been considered for model evaluation. The work performed during the study was built on keywords of the vernacular language Wolof, based on many open source software toolkits, particularly HTK (HMM ToolKit). Much research have been developed in this area; our goal is also to find solution for an innovative approach to Speech Recognition to facilitate access to information and technology to illiterate persons, to build a phonetic crowdsourcing based on acoustic and linguistic features of local languages.