A configurable distributed speech recognition system, Biennial on DSP for in-Vehicle and Mobile Systems (original) (raw)

This paper presents a configurable distributed speech recognition (DSR) system designed to enhance the speech-driven application experience on mobile devices. The proposed system leverages a client-server architecture where the client focuses on feature extraction, while the server executes the more computationally intensive recognition decoding tasks. Key advancements include the integration of ETSI-DSR standards and a modified AFE to improve noise robustness. Evaluation results demonstrate the system's efficiency, low resource consumption, and adaptability to various recognition tasks, addressing previous limitations in real-life DSR implementations.