WAVELET-BASED ENERGY BINNING CEPSTRAL FEATURES FOR AUTOMATIC SPEECH RECOGNITION

摘要

Systems and methods for processing acoustic speech signals which utilize thewavelettransform (and alternatively, the Fourier transform) as a fundamental tool.The method essentiallyinvolves "synchrosqueezing" spectral component data obtained by performing awavelet transform(or Fourier transform) on digitized speech signals. In one aspect, spectralcomponents of thesynchrosqueezed plane are dynamically tracked via a K-means clusteringalgorithm. The amplitude,frequency and bandwidth of each of the; components are, thus, extracted. Thecepstrum generatedfrom this information is referred to as "K-mean Wastrum." In another aspect,the result of theK-mean clustering process is further processed to limit the set of primarycomponents to formants.The resulting features are referred to as "formant-based wastrum." Formantsare interpolated inunvoiced regions and the contribution of unvoiced turbulent part of thespectrum are added. Thismethod requires adequate formant tracking. The resulting robust formantextraction has a numberof applications in speech processing and analysis including vocal tractnormalization.