ITB Journal

The development of speech recognition engines has traditionally been the territory of low-level development languages such as C. Until recently Java may not have been considered a candidate language for the development of such a speech engine, due to its security restrictions which limited its sound processing features. The release of the Java Sound API as part of the Java Media Framework and the subsequent integration of the Sound API into the standard Java development kit provides the necessary sound processing tools to Java to perform speech recognition. This paper documents our development of a speech recognition engine using the Java programming language. We discuss the theory of speech recognition engines using stochastic techniques such as Hidden Markov Models that we employ in our Java based implementation of speech signal processing algorithms like Fast Fourier Transform and Mel Frequency Cepstral Coefficients. Furthermore we describe our design goal and implementation of a distributed speech engine component which provides a client server approach to speech recognition. The distributed architecture allows us to deliver speech recognition technology and applications to a range of low powered devices such as PDAs and mobile phones which otherwise may not have the requisite computing power onboard to perform speech recognition .