Experiments on front-end techniques and segmentation model for robust Indian Language speech recognizer

Author

Sriranjani, R. ; Karthick, B. Murali ; Umesh, S.

Author_Institution

Dept. of Appl. Mech., Indian Inst. of Technol., Chennai, Chennai, India

fYear

2014

fDate

Feb. 28 2014-March 2 2014

Firstpage

1

Lastpage

6

Abstract

Recent contributions in the area of Automatic Speech Recognition (ASR) for Indian Languages has been increased. This paper serves as a comprehensive study of different feature extraction methods namely MFCC, PLP, RASTA-PLP and PNCC. An attempt to find out which of these front end techniques performs better for real world Indian Language data is analyzed experimentally. Then, an isolated word recognizer is built for three Indian languages (i.e., Tamil, Assamese and Bengali) under real world conditions and investigates the importance of handling long silence using segmentation method. The experimental analysis shows that PNCC provides better performance for clean data whereas MFCC shows improved performance in case of multi-condition speech data.

Keywords

feature extraction; natural language processing; speech recognition; automatic speech recognition; feature extraction methods; front-end techniques; isolated word recognizer; multicondition speech data; robust language speech recognizer; segmentation model; Data models; Feature extraction; Hidden Markov models; Mel frequency cepstral coefficient; Noise; Speech; Feature extraction; Noise robustness; Segmentation; Silence handling; Speech recognition; comparison of front-end techniques; real world speech;

fLanguage

English

Publisher

ieee

Conference_Titel

Communications (NCC), 2014 Twentieth National Conference on

Conference_Location

Kanpur

Type

conf

DOI

10.1109/NCC.2014.6811284

Filename

6811284