Analysis of Stop Consonants in Indian Languages Using Excitation Source Information in Speech Signal

In this paper we propose excitation-based features for extracting information about the manner of articulation for stop consonants. The excitation-based features are derived from very low frequency information in the signal and also from the normalized error computed from the linear prediction residual. The proposed zero-frequency filtered signal brings out the region of glottal activity during excitation. Likewise, the normalized error helps to distinguish regions of noise and pure voicing. These nonspectral methods of analysis of stop consonants seem to provide additional and some better features over the features derived from the traditional methods based on short-time spectrum analysis.