Application of Eαnets to feature recognition of articulation manner in knowledge-based automatic speech recognition

  • Authors:

  • Sabato M. Siniscalchi;Jinyu Li;Giovanni Pilato;Giorgio Vassallo;Mark A. Clements;Antonio Gentile;Filippo Sorbello

  • Affiliations:

  • Center for Signal and Image Processing, School of Electrical and Computer Engineering, Georgia Institute of Technology, Atlanta, Georgia, United States of America;Center for Signal and Image Processing, School of Electrical and Computer Engineering, Georgia Institute of Technology, Atlanta, Georgia, United States of America;Istituto di CAlcolo e Reti ad alte prestazioni, Italian National Research Council, Palermo, Italy;Dipartimento di Ingegneria Informatica, Universita' degli studi di Palermo, Palermo, Italy;Center for Signal and Image Processing, School of Electrical and Computer Engineering, Georgia Institute of Technology, Atlanta, Georgia, United States of America;Dipartimento di Ingegneria Informatica, Universita' degli studi di Palermo, Palermo, Italy;Dipartimento di Ingegneria Informatica, Universita' degli studi di Palermo, Palermo, Italy

  • Venue:
  • WIRN'05 Proceedings of the 16th Italian conference on Neural Nets
  • Year:
  • 2005

Quantified Score

Hi-index 0.00

Visualization

Abstract

Speech recognition has become common in many application domains. Incorporating acoustic-phonetic knowledge into Automatic Speech Recognition (ASR) systems design has been proven a viable approach to rise ASR accuracy. Manner of articulation attributes such as vowel, stop, fricative, approximant, nasal, and silence are examples of such knowledge. Neural networks have already been used successfully as detectors for manner of articulation attributes starting from representations of speech signal frames. In this paper, a set of six detectors for the above mentioned attributes is designed based on the E-αNet model of neural networks. This model was chosen for its capability to learn hidden activation functions that results in better generalization properties. Experimental set-up and results are presented that show an average 3.5% improvement over a baseline neural network implementation.