Eigenstructure based pitch estimation in speech processing applications

The author continues the development of eigenstructure-based techniques for pitch detection and estimation in speech processing. The cepstral model and method are reviewed, and a noisy cepstral signal model is described. The effect of performing eigenanalysis on a logarithmically transformed sinusoid is analyzed. The noisy model and eigenanalysis results are combined. Some preliminary evaluations using speech data from the Texas Instruments Long Distance Telephone Pitch Detection database are given. >