Pitch estimation and a measure of voicing from pseudo-spectra

In the sinusoidal model of speech, the analysis and synthesis is carried out using a superposition of sinusoidal components whose frequencies, amplitudes, and phases are to be estimated. We apply subspace and linear prediction methods of frequency estimation to speech waveforms using this model. These techniques yield a measure of the degree of voicing. They also lead to a new robust and reliable pitch estimation algorithm.