A Voice Controlled UHF Radio

Abstract

Nowadays, more and more devices attempt to improve user interaction including the use of speech recognition. This project responds to the need of a company for developing an efficient system for voice control of an operative UHF radio in sporting environments. The work, conducted at Chalmers has been divided into two main parts. First, a theoretical study and analysis of the “state of art” in voice control algorithms has been performed. The initial objective of this part was to familiarize and understand how diverse systems and algorithms in voice recognition work, even testing and modifying some of them. In the second phase, the thesis has been focused on developing and testing a demonstrator device with the concepts and ideas learned from the theory. An Android device recognizes specific voice commands and sends the orders to the UHF radio by a combination of serial and TCP/IP communications. For control with maximum accuracy and minimum power consumption, Hidden Markov Models (HMMs) and Dynamic Time Warping (DTW) algorithms for voice recognition have been studied, optimized and implemented through several applications in Java. The thesis concludes with several reflections about how the thesis has been developed and it introduces some ideas and recommendation to improve the system in future researches.