Implementation and Performance Evaluation of Speaker Adaptive Continuous Hindi ASR using Tri- phone based Acoustic Modelling

Ankit Kuamr¹, Mohit Dua²

^1,2National Institute of Technology Kurukshetra, India

Abstract

Speech interface to computer is the next big step that computer science needs to take for the general users. Speaking in our native language is a natural and effortless task which carried out with great speed and ease. Speech recognition will play an important role in taking technology to common man. The need is not only for speech interface but speech interface in local language like Hindi. In this paper, we evaluate the performance of Hindi Automatic Speech Recognition (ASR) by using two most popular feature extraction techniques namely Mel Frequency Cepstral Coefficient (MFCC) and Perceptual Linear Prediction (PLP) at front- end of ASR. The Hidden Markov Model (HMM) was used at back- end of an ASR for Hindi language. The proposed system has been implemented for the continuous Hindi speech using tri-phone based acoustic modelling. For speaker adaptation Maximum Likelihood Linear Regression (MLLR) technique has been implemented in this paper. All experiments were done by using HTK 3.4.1 (Hidden Markov Toolkit) on Linux environment (Ubuntu 12.05).

Keywords

• Hindi Speech Recognition, Speaker Adaptation, automatic speech recognition, HMM, MFCC, MLLR

Author's Profile

Author profile can be generated and linked through our partners World Book of Researchers. To include your profile online Click Here. After it is approved, please email to edlib @ asdf.res.in to create a link with all the papers.

• Ankit Kuamr : Profile
• Mohit Dua : Profile

e-AID

Cite this Article as Follows

Ankit Kuamr, Mohit Dua. Implementation and Performance Evaluation of Speaker Adaptive Continuous Hindi ASR using Tri- phone based Acoustic Modelling. International Conference on Human Machine Interaction 2014. Vol. 1. Chennai: Association of Scientists, Developers and Faculties, 2014. 1-6. Print.