Vocal tract shape estimation using three noninvasive transducers
Russel Long, Brad H. Story, Ingo R. Titze · The Journal of the Acoustical Society of America · 1998
Sentence-level speech simulation requires a description of vocal tract shape parameters as a function of time. Vowel-like configurations can often be derived from a microphone signal; however, the influence of consonants may be more easily detected with the use of additional signals. The electroglottograph (EGG) and an articulatory transducer [McGarr and Lofqvist, J. Acoust. Soc. Am. 72, 34–42 (1982)] are employed to increase the accuracy of vocal tract shape estimation. The microphone signal is analyzed on a frame-by-frame basis for spectral features, energy, and zero-crossing rate. The EGG provides information regarding voicing onset and offset, and the articulatory signal is used to indicate closure and lip rounding. These data are combined to select an appropriate area function for each frame of the sentence. The resulting temporal map of area functions is interpolated and sent to a wave reflection simulator to reconstruct the speech. [Work supported by NIH Grant No. DCO2532.]