Vocal tract spectrum estimation device, method, and program

摘要

PROBLEM TO BE SOLVED: To estimate a vocal tract spectrum precisely by using a speech signal.SOLUTION: A time frequency expansion unit 24 inputs time series data of a speech signal, and then outputs an observation spectrogram. A parameter-estimating unit 36 estimates a mixed Gauss function model, and further estimates: K formant frequency sets and a coupling factor of each time; a formant intensity with regard to the K formant frequency sets; and a formant dispersion and a coupling factor of each time with regard to the K formant frequency sets, to reduce a standard for showing the magnitude of error with the observation spectrogram. The mixed Gauss function model is obtained when, with regard to each Gauss distribution function corresponding to each peak: an average is expressed by the K formant frequency sets and the coupling factor; weight is expressed by the formant intensity and the coupling factor with regard to the K formant frequency sets; and dispersion is expressed by the formant dispersion and the coupling factor with regard to the K formant frequency sets.SELECTED DRAWING: Figure 5