Eight‐Parameter Speech Synthesis

PAT, the Parametric Artificial Talking device (described in abstract G1 by Walter Lawrence) recently has been modified to synthesize speech using the following eight parameters: larynx frequency, larynx amplitude, three formants. frequency of the hiss filter, amplitude of hiss in the hiss filter, and amplitude of hiss in the second and third formants. By using these parameters speech has been synthesized both from spectrographic analysis and from minimal rules similar to those presented in earlier papers [Ingemann, J. Acoust Soc. Am. 29. 1255 (1957); A. M. Liberman, F. Ingemann, L. Lisker, P. Delattre, and F. S. Cooper, J. Acoust. Soc. Am. 31, 1490–1499 (1959)]. The rules originally designed for the Pattern Playback at Haskins Laboratories had to be somewhat modified and further simplified to remain within the eight parameters of PAT, but nevertheless intelligible speech has been produced. The types of modifications necessary will be discussed, and recordings of speech synthesized from analysis and by rule will be played.