A comparison of distributed and network speech recognition for mobile communication systems

In this paper, we compare the conventional Network Speech Recognition (NSR) and the newly established Distributed Speech Recognition (DSR) concepts for mobile communications. These implementation approaches to Automatic Speech Recognition (ASR) are analyzed from three aspects. First, the effect on the speech recognition accuracy of ASR systems with various complexity. Second, usability in different operating environments (environmental noises). Finally, the resilience to erroneous transmission (radio channel). Our experimental results show that DSR reduces the error rate of ASR systems by 1725% on average compared to NSR, depending on the recognition task. The error resilience of DSR provides up to 37% error rate reduction over NSR under severe channel error conditions. In non-stationary environmental noise, DSR may outperform NSR by as much as 41% in terms of error-rate reduction. Based on these results and considering the low complexity implementation of DSR in the mobile terminals, it is concluded that DSR provides a viable, robust and economical alternative to the traditional NSR approach, especially in real-life mobile operating environments.