ASR Assisted Speech Transcription in Sensor Networks

Abstract

Wireless sensor networks are becoming increasingly popular and are widely used in many applications. Sensors can be used for environment and habitat monitoring, healthcare applications, home automation and traffic control. Many of these applications require sensors with speech recording/transmitting capabilities. In such cases, bandwidth usage could be significantly reduced if speech is transmitted as a sequence of automatically recognized words. In this work we propose a system which incorporates speech recognition capable sensor nodes and a central unit operated by a human transcriber. Our approach is to combine the two estimates (from human and machine speech recognizers) based on their reliability in order to achieve an improved estimate.