Thesis

Speech recognition techniques using linear prediction

Speech recognition is cast as a problem in general recognition theory. Linear prediction methods are presented as ways of characterizing spectral information in the speech waveform for use in speech recognition systems. Linear prediction equations are developed from an all-pole speech production model. Two variations of linear prediction are discussed: the autocorrelation method and the covariance method. Fortran programs for analyzing speech data using these techniques are given.

Relationships

Items