Despite big technological advancements, we are still bound to use keyboard and mouse as default computer peripheral. When using mobile devices, the scope of possible interactions with a device quickly grows. Many mobile device users would like to use voice commands for both simple (voice dial), as well as for demanding tasks such as web search. Usually, it takes a number of learning examples on which speech recognizer learns and improves. In this thesis, we are trying to avoid this necessity. The main goal of this thesis is the development of a mobile application that recognizes speech without a demand for a substantial number of learning examples and which will operate completely independently, without active internet connection. The method used for speech recognition system in this thesis is well-known Dynamic Time Warping method. Analysis, where we compared accuracy and speed of recognition, was carried out. Based on the results of the analysis we have successfully proposed a few steps to optimize the speed of our speech recognition application and preserve its accuracy.
|