Comparison of acoustical models of GMM-HMM based for speech recognition in Hindi using PocketSphinx

Abstract

Automatic Speech recognition (ASR) is widely gaining momentum worldwide, to be used as a part of Human Computer Interface and also in a wide variety of commercial applications. In Indian context, commercial applications using automatic speech recognition are still in the evolving process. This paper describes the acoustic models that have been cross language adapted for speech recognition in Hindi using CMU’s PocketSphinx. A database of 177 words in Hindi is prepared, along with transcription and dictionary. Two approaches for developing acoustical model for the speech recognition has been discussed in this paper. In the first approach, English acoustical model has been cross-language adapted to Hindi. In the second approach different acoustical models -continuous, semi continuous and phonetically tied models have been trained. GMM-HMM is used for acoustical modeling and language modeling.