Keyword Recognition using Perceptually Significant Features

Padmalochini Umakanthan and Kaliappan Gopalan


Spectrally significant energy, Speech recognition, Cepstral features, DTW


This paper presents the results of a speech recognition experiment conducted to recognize certain keywords using perceptually significant energy features. Spectral components that are above global masking threshold in each of the 17 bands in each frame of a word utterance form the features. Dynamic time warping was used for feature matching. For validation of the technique results were compared with those obtained using cepstral features. Preliminary results indicate the viability of using the perceptually significant features as an augmented feature set in keyword spotting.

Important Links:

Go Back