Real-time synthesis of optical lip shapes from vowel sounds

Publisher Website
Google Scholar
Abstract
An analog circuit has been constructed to convert speech (vowels) to Lissajous-type oscilloscope patterns in real time. The method is like that described by Boston [Brit. J. Audiol. 7, 95–101 (1973)], but includes circuit modifications to improve naturalness of lip outline and articulatory change. The speech signal receives preemphasis resembling ear canal resonance and is split into two channels. Two voltages, one proportional to 200–1000 Hz energy (F1) and the other proportional to 200–3000 Hz energy (F2 in front vowels, F1 in back vowels), are generated to control a multiplier and divider, respectively, which vary levels of sinusoidal signals to form the height and width of the lip outline. Resulting patterns resemble lip shapes of vowels in natural speech. Vowel-confusion matrices obtained by lipreading the oscilloscope display (with speech input) are similar to those obtained by lipreading the talker’s face directly.

This publication has 0 references indexed in Scilit: