A better LASS
D. Lloyd Rice · The Journal of the Acoustical Society of America · 1976
The first Line Analog Speech Synthesizer (LASS) system used at UCLA modeled a vocal tract consisting of 18 sections, each with an area and an acoustic loss factor that can be specified independently. For computational reasons each section had to be the same length, which was obviously a disadvantage. Another weakness was that the radiation impedance at the lips was modeled in a somewhat simplistic manner. Both these disadvantages have now been overcome, without losing the advantage of a model in which the losses can be specified at each point along the tract. A curve is interpolated between specified points along the tongue, converted to areas and smoothed to determine 18 evenly spaced sections. The losses at the lips are now modeled by a realistic resistive-inductive load. The formants calculated using this model have been compared with those calculated by the previous model, and with those calculated by a model described by Fant. There are still differences between calculated and observed formant frequencies. We do not know how to decide whether the difficulty in reproducing observed formant frequencies is due to inaccurate estimates of the acoustic losses along the tract or errors in the conversion from sagittal dimensions to cross-sectional areas, or some other factors.