The speech analysis technique presented in this paper is based on a novel pole-focusing mechanism. This method of analysis is shown to greatly improve the noise immunity in the extraction of resonance features from linear prediction spectra. The detection of very low-intensity spectral features is enhanced compared to standard LPC analysis, and superior separation is provided of features which tend to merge in the time-frequency plane. The technique employs off-axis spectral estimation coupled with a progressive increase in the LPC model order. Contrary to the standard LPC analysis, preemphasis is found to degrade the performance of the pole-focusing techniaue. Finally, the pole focusing technique appears to be less sensitive to choice of the vocal tract model order.
展开▼