Speech synthesis at Haskins Laboratories: From the pattern playback to articulatory synthesis

While the creation of devices that talk or replicate aspects of speech stretches back over hundreds of years, the use of speech synthesis for systematic research was accelerated by the development of the Pattern Playback at Haskins. This device simulated the acoustics of speech by copying or altering spectrograms. Despite its limitations, it supported foundational work in the 1950s on the underlying critical “cues” for perception. Later developments at Haskins and other locations allowed for completely digital control (including the Digital Pattern Playback). Synthesis by rule converted text into understandable (if somewhat tiresome) speech. This work had a revolutionary effect on speech science and was fundamental to the development of modern speech synthesis, the development of reading machines for the blind, and the study of speech recognition. It also raised a serious consideration of speech production in the synthesis process. Using a vocal tract model developed at Bell Laboratories by Mermelstein, Coker, and others, Rubin and colleagues developed the first articulatory synthesizer (ASY) with articulation as the input. ASY helped shape the development of articulatory phonology (Browman and Goldstein). Later, a configurable, 3-D model of the vocal tract (CASY) permitted replication of MRI images of articulation.