Feature Compensation Combining SNR-Dependent Feature Reconstruction and Class Histogram Equalization

  • Suh, Young-Joo (School of Engineering, Information and Communications University) ;
  • Kim, Hoi-Rin (School of Engineering, Information and Communications University)
  • Received : 2008.05.14
  • Accepted : 2008.06.23
  • Published : 2008.10.31

Abstract

In this letter, we propose a new histogram equalization technique for feature compensation in speech recognition under noisy environments. The proposed approach combines a signal-to-noise-ratio-dependent feature reconstruction method and the class histogram equalization technique to effectively reduce the acoustic mismatch present in noisy speech features. Experimental results from the Aurora 2 task confirm the superiority of the proposed approach for acoustic feature compensation.

Acknowledgement

Grant : Development of Large Vocabulary/Interactive Distributed/Embedded VUI for New Growth Engine Industries

Supported by : IITA