Research On The Distributed Speech Recognition Application On Mobile Phone

As the booming of mobile communication, voice recognition technology has found its more and more important role in mobile phones. The large computation request for Voice Recognition algorithm is the major restriction for mobile device, whose computation capability is relatively limited. Distributed Speech Recognition (DSR) provides a new solution for this by splitting Voice Recognition algorithm into front end and back end. As DSR front end, mobile device only executes voice detection and parameter extraction functions. While in back end, DSR server performs the complicated recognition work. Front end and back end communicates with each other through the network. As the computer network becomes popular and can provide much more bandwidth, DSR is getting more and more focus in Voice Recognition area.In this thesis, the author has in depth research of the principles and each components of DSR, implements DSR architecture and analyzes its performance. Voice parameter extraction is the major task of the DSR front end. In this thesis, some algorithms of MFCC abstraction and linear prediction are analyzed. Experiments are designed to compare the algorithm's performance in different noise strength environment. As a result, Spectral Subtraction is used to reduce the system noise and the overall performance is improved.An actual DSR system has been implemented in Motorola i605 radio. Through this practice, the author summarizes the methodology of DSR system design in mobile devices, in terms of architecture design, detailed implementation, as well as finishing the software development in the application layer.