Robust Speech Recognition in Wireless Communication

This project aims at potential solutions for the robustness of Distributed Speech Recognition (DSR) which probably runs over a wireless network. The popular speech recognition architecture DSR divides a speech recognizer into a client and a server parts , which brings not only much flexibility but simultaneously some new problems. In this context, acoustic noises from environments and packet loss during network transmission will largely degrade the recognition performance. However, the client often runs in a resource limited mobile device, which could not endure the heavy computational costs brought by commonly used robust methods. Therefore, effective and efficient methods should be developed. This project mainly focuses on the problems arising from environmental noise and incorporates results from ongoing research within the SMC group on alleviating transmission errors in an attempt to propose innovative and efficient solutions for both the client and the server modules.