Hierarchical automatic speech recognition powered by data infrastructure

Abstract

Automatic Speech Recognition (ASR) has evolved remarkably over the years and is expected to become a primary form of input to mobile devices including smartphones and wearables. Most large-scale mobile platforms perform speech recognition in the cloud today. There are both advantages and disadvantages to this Cloud-based ASR (Cloud-ASR) approach. Cloud-ASR approach allows for a context oriented humancomputer- interaction using speech rather than a mere speech-totext translation. A Cloud-ASR also has disadvantages such as interruption of the speech service when there is no access to the Cloud-ASR, and also the energy consumption for radio communications, which can drain a mobile battery sooner. We propose the usage of Hierarchical Speech Recognizer (HSR) as an alternative approach to overcome the shortcomings of the Cloud-ASR approach. In the HSR approach, mobile devices perform "selective speech recognition" by themselves as much as possible without contacting an external cloud-based ASR service. In this demonstration, we show our proof-of-concept HSR along with its feasibility and advantages.