Opportunities & Challenges In Automatic Speech Recognition

Automatic speech recognition enables a wide range of current and emerging applications such as automatic transcription, multimedia content analysis, and natural human-computer interfaces. This paper provides a glimpse of the opportunities and challenges that parallelism provides for automatic speech recognition and related application research from the point of view of speech researchers. The increasing parallelism in computing platforms opens three major possibilities for speech recognition systems: improving recognition accuracy in non-ideal, everyday noisy environments; increasing recognition throughput in batch processing of speech data; and reducing recognition latency in realtime usage scenarios. This paper describes technical challenges, approaches taken, and possible directions for future research to guide the design of efficient parallel software and hardware infrastructures.