Speech Activity Detection for Deaf People: Evaluation on the Developed Smart Solution Prototype

Abstract

This research constitutes a relatively new approach by developing a smart solution which has emerged from the research activity using at first Google Glass and usable speech detection services. The authors conducted in the last year a series of developing, testing and evaluating the prototype results in order to decide, which service provides better results than the third-party speech detection service like Google Speech API or IBM Watson Speech To Text. This finding should significantly help the authors during the data evaluation and testing in developed smart solution. The basic idea is that authors have already developed a functional basic solution—a prototype. This solution was properly working and usable, but there are still some disadvantages to be improved. In order to accomplish the best results possible, the authors have added another element to their solution. A challenging problem which arises in this domain is concerned with significant data savings, server load, detection quality, and again opens a space for further improvements, such as following research and testing. This element is part of the statistical analysis and it is called Hidden Markov Model, which is used for speech recognition applications for last twenty years. The authors examined and studied many different articles and scientific sources in order to find the best solution for higher efficiency of speech recognition usable in their developed prototype (and for this article).