Pervasive speech recognition

Abstract

Various aspects of the embedded automatic speech recognition (ASR) system are discussed. ASR uses a pheneme-based approach for speech modeling and the quality of speech an ASR system receives depends largely on the microphone's type and location. ASR consists of transforming the raw speech input into a hypothesized sequence of words. It is observed that in one instance of an ASR system, the front-end module inputs 16 kHz speech samples coded onto 16 bits.