Offline Sindhi Speech Recognition

Abstract

Motivation behind this research effort is to fill up the gap between the computer and its related technologies and the people of rural Sindh, by enabling them to use Sindhi language in their computers, mobiles etc. The Sindhi language is the most commonly used language by Sindh province people, particularly people living in rural areas of Sindh. This language is also spoken in many areas of India. While there exists a number of speech recognition systems, which have been developed for different languages, literature suggests there is need for speech recognition system in local Sindhi language. We have designed and developed a speech recognition system in Sindhi language, which is based on Artificial Neural Network (ANN) and Hidden Markov Model (HMM). The main purpose behind using ANN is that has an adaptive nature which means it has the capability to learn. In this paper, we present and discuss various aspects of the proposed Sindhi speech recognition system including fundamental model of a speech recognition trainer and its different modules. It also describes how to develop dictionary for Sindhi language and how to train a module for Automatic Speech Recognition (ASR). We also present and discuss the results of experiments conducted to evaluate the system.