A Framework for Multiple Choice Multilingual Translation System Using Hidden Markov Model and Viterbi Algorithm

Abstract

In the multilingual World, majority of languages are in parallel to each other, which make communication among different speakers difficult and burdensome. Most of the existing approaches to language translation focuses on either speech-to-text, text-to-speech, speech-to-speech or text-to-text, but do not consider user’s preferences. In this paper, we present a framework for multiple choice multilingual translation system to convert the input English speech signals, text and printed text into Speech and/or text output for users in either Hausa, Igbo or Yoruba. Intuitively, the system consists of four modules, which include text extraction, speech recognition, text translation and speech synthesis modules. We used Mel Frequency Cepstral Coefficients (MFCC) to extract features from the speech signals of spoken words. Furthermore, we used Hidden Markov Model to train and test the audio files to get the recognized spoken word. The Viterbi Algorithm was used to get the most likely path and word combinations. For scanned images and printed documents, Optical Character Recognition was used for text extraction.