Joint estimation of the voiced component and spectrum of a speech signal

We discuss a method for extracting simultaneously both the formant information and the voiced components of a speech signal. The formant information is assumed to reside in the filter which models the vocal tract, and the voiced components are assumed to be the periodic components of the vocal tract excitation signal. These two distinct pieces of information can be used in many areas of speech processing, including low bit rate speech coding, speech recognition, and speech enhancement.