An efficient distributed speech processing in noisy mobile communications

Abstract

Mobile communications are greatly influenced by environmental noise that may cause a significant deterioration in automatic speech recognition (ASR) systems performance. In this paper, we present a new framework integrating a noise-robust front-end in distributed speech recognition (DSR) systems. Using the Aurora-2 speech database, the authors evaluate the development of an additional feature set for Mel-frequency-based European Telecommunications Standards Institute advanced front-end (ETSI-AFE) which we refer to as power-normalized cepstral coefficients (PNCCs). The experimental results show that, the proposed approach achieves a significant improvement in word recognition accuracy compared to the current ETSI-AFE deployed by the DSR technology on today's mobile phones.