Displaying speech as vocal tract area function pictures

Various linear predictive models of the speech process enable estimates of the vocal tract area function (VTAF) to be obtained throughout a speech utterance. This paper examines the benefits and difficulties of displaying the VTAFs as intensity‐modulated digital pictures. In this display, distance from the glottis is plotted along the vertical axis, time is plotted as the horizontal axis, and the area is shown as intensity. The uses proposed for this display include speech and phonetic studies, automatic recognition algorithm testing and invariant feature enhancement and extraction. The major problem associated with the VTAF picture is caused by the breakdown of linear prediction models for nonvoiced speech, particularly the fricatives. A pattern recognition algorithm to detect fricatives and compensate for this weakness is described. [Work supported by DRCS.]