Java (J2ME) Front-End for Distributed Speech Recognition

In this paper we introduce a heretofore unheard-of implementation of the front-end for distributed speech recognition (DSR) systems using Java platform, micro edition (J2ME) framework. The developed floating and fixed point versions of the front-end deliver features conforming the ETSI standard ES201108, whereas our J2ME fixed-point front-end provides 25% improvement in word error rate compared to the standard one. The presented results of the extensive experiments with different mobile devices show that feature extraction and compression within real time is possible on today´s mobile phones.