Design of a Limited Speech Recognition System for use in a Braille Teaching Device

Abstract

<p>The report here submitted defines the scope and content of the Electrical and Biomedical Engineering Capstone Project as submitted by Brett Lindsay. This project involved the creation of a limited Speech Recognition system for use in a Braille Teaching Device. The greater project (that of the Braille Teaching Device) was completed in tandem with Messrs. Chris Agam and Jonathon Hernandez. It was felt that the Speech Recognition component would be a valuable addition to the project due to the nature of a teaching device for use by the visually impaired (who would need an assistant to use said device). The Speech Recognition system was creating by breaking the problem into four subsections: the collection of data upon call by the teaching program, the manipulation of data, the recognition algorithms to categorize said data, and the passing of results back to the teaching program. For the recognition block, the relatively simple method of Dynamic Time Warping was chosen over more complex options such as Hidden Markov Models or Neural Networks. This method presented some problems as documented, specifically a tendency to favour letters with larger file sizes (such as 'w'). The Speech Recognition system created during the course of this project failed to deliver on the wanted efficiency of 60 % and low as possible false positives. While the Speech Recognition presented is viable, the effectiveness is below that which can be found in market for comparable price.</p>