Untitled Document
You are from : ( )  
Untitled Document
Untitled Document

International Journal of Information Technology & Computer Science ( IJITCS )

Abstract :

This paper presents speech processing and recognition system for Myanmar continuous language. The speech segmentation methods are based on two simple speech features, namely time domain features and frequency features. To detect the word boundaries, dynamic thresholding method is applied. Then, important features of speech features are extracted by LPC (linear predictive coding) and GTCC (gamma tone cepstral coefficient) approach.  K-means is used in feature clustering and HMM is used in recognition process.   All the algorithms used in this work are implemented in Matlab. The system obtained the average word error rate of 4.16 and 0.479 with LPC and GTCC feature extraction techniques respectively.

Keywords :

: automatic speech recognition; GTCC; LPC; HMM

References :

  1. O. Cheng, W. Abdulla,    Z. Salcic, “Performance Evaluation of Front-end Processing for Speech Recognition “, School of Engineering Report No. 621
  2. L. Rabiner, Fellow, IEEE  ‘A Tutorial On Hidden Markov Model And Selected Applications In Speech Recognition, Proceedings Of The IEEE, Vol. 77, No.  2, February 1989
  3. http://www.cs.brown.edu/research/ai/dynamics/tutorial/Documents/HiddenMarkovModels.html
  4. T. Giannakopoulos, “Study  and  application  of  acoustic  information  for the  detection  of  harmful  content  and  fusion  with  visual  information” Ph.D. dissertation, Dept.of Informatics  and  Telecommunications, University of Athens, Greece, 2009.
  5. T. Giannakopoulos, A. Pikrakis and S. Theodoridis “A  Novel  Efficient Approach  for  Audio  Segmentation”,  Proceedings  of  the  19th International  Conference  on  Pattern  Recognition  (ICPR2008), December 8-11 2008, Tampa, Florida, USA.

Untitled Document
Untitled Document
  Copyright © 2013 IJITCS.  All rights reserved. IISRC® is a registered trademark of IJITCS Properties.