Hindi Digit Recognition Using KALDI ASR toolkit

  • Prashant Upadhyaya, Sheetal Upadhyaya, Omar Farooq, Paras Chawla

Abstract

In this paper, Hindi digit recognition using KaldiASR toolkit is presented. For digit recognition, the
Mel Frequency Cepstral Coefficients (MFCC) are extracted from speech samples. The performance of
the automatic digit recognition model was evaluated for both Speaker Independent (SI) and Speaker
Dependent (SD) model. For comparative performance, Hindi digit recognition was evaluated on the
Matlab using the LDA (Linear Discriminant Analysis) classifier for both SI and SD model using the
same MFCC features. Maximum improvement in the accuracy for the Hindi digit recognition using
Kaldi ASR for SI and SD was 28.43% and 21.94% respectively with respect to LDA classifier. Finally,
it was experimentally demonstrated, that the Hindi digit recognition accuracy can be increased with
the help of Kaldi ASR toolkit.

Published
2020-05-20
How to Cite
Prashant Upadhyaya, Sheetal Upadhyaya, Omar Farooq, Paras Chawla. (2020). Hindi Digit Recognition Using KALDI ASR toolkit. International Journal of Advanced Science and Technology, 29(10s), 2127-2131. Retrieved from http://sersc.org/journals/index.php/IJAST/article/view/16823
Section
Articles