Reusing automatic speech recognition platform for resource deficient languages

摘要

Indian subcontinent is diverse with several spoken languages in use and a wide range in literacy levels of the inhabitants. In this scenario, speech recognition based solution fits best to enable all strata of population access information. While speech recognition based solutions in the Western countries have moved into commercial use, they are far from being used in Indian scenario, mainly due to lack of language resources. Automatic Speech Recognition (ASR) for Indian languages, including Indian English for India specific conditions is still maturing. In this paper, we show how existing resources can be effectively reused, to build usable speech based solution for resource deficient languages. We specifically employ PocketSphinx and Google Speech API to evaluate our ideas.