The performance evaluation of distributed speech recognition for Chinese digits

Abstract

Distributed speech recognition (DSR) based on client-server mode means that the recognizer front-end is located in the terminal and is connected over a data network to a remote back-end recognition server. In this paper, we evaluate the distributed speech system in the client-server framework based on the ETSI AURORA platform for Chinese digit recognition and compare the performance of client-server mode with the server-only mode where speech is transmitted by using the GSM AMR encoder and all recognition processing is done at the server side. The recognition result shows that without consideration of channel error the DSR system based on client-server mode can achieve satisfactory recognition accuracy and the lowest hit rate in comparison with sever-only mode.