Speech Recognition for Keyword Spotting using a Set of Modulation Based Features – Preliminary Results

We present the preliminary results of applying a set of parameters of the AM-FM model for recognizing word utterances. By acquiring modulation based parameters from the amplitude envelope (AE) and the instantaneous frequency – both obtained by demodulating at four selected center frequencies – a compact feature set is created for each frame of a word utterance. Applying a dynamic time warping of features, a dissimilarity measure between an unknown and one of several reference utterances is obtained to detect the presence of a keyword in a continuous stream of speech. A feature set consisting of the peak frequencies in AE and weighted formants, among others, shows an overall recognition score of 75 percent or higher – depending on the analysis frequencies used – for an extracted set of word utterances. The low false positive and false negative scores suggest the viability of modulation based parameters for building a keyword spotting system.