A model of acoustic interspeaker variability based on the concept of formant-cavity affiliation.

摘要: A method is proposed to model the interspeaker variability of formant patterns for oral vowels. It assumed that this originates in differences existing among speakers respective lengths their front and back vocal-tract cavities. In order characterize, from spectral description acoustic speech signal, these between speakers, each interpreted, according concept formant-cavity affiliation, as a resonance specific cavity. Its frequency can thus be directly related corresponding cavity length, transformation speaker B on basis ratios formants same resonances. minimize number sounds recorded carry out transformation, are exactly computed only three extreme cardinal vowels [i, a, u] they approximated remaining through an interpolation function. The evaluated its capacity transform (F1,F2) eight pronounced by five male into generated articulatory vocal tract. resulting compared those provided normalization techniques published literature. found efficient, but limitations also observed discussed. These associated with affiliation itself or possible influence speaker-specific geometry cross-sectional direction, which might not have taken account.