Early DIVA Simulations

This page presents videos of DIVA model simulations from the 1990’s.

The first video illustrates two different types of babbling: (1) completely random articulator movements, and (2) reduplicated babbling, created by moving the jaw up and down cyclically, with tongue movements only occurring between the bi-syllabic utterances. Note how the addition of  cyclic jaw movements (at the right frequency) makes the productions sound much more “speech-like”. Babies typically undergo a reduplicated babbling stage at around 8-10 months of age, at which time their utterances start to sound distinctly more like words such as “mama” and “dada”.

In the next video, the model (sporting a receding hairline for some reason) produces the vowels in “bet”, “beet”, “bat”, “but”, and “boot”, first under normal conditions, and then with the jaw immobilized, illustrating its ability to compensate for constraints or perturbations to the speech articulators.

The next video illustrates the model producing the word “baby” three times, with an increasingly refined auditory target each time. Note how the word becomes more clear and the articulator movements more precise as the auditory target improves.

The next video illustrates the model’s first six attempts to produce the utterance “good doggie” after being given an auditory target for the utterance. The first attempt is primarily under auditory feedback control and sounds “sloppy”. Each subsequent attempt relies more heavily on learned feedforward commands, resulting in clearer and clearer productions.

In the next video, the model produces “good doggie”, with palatal contact information, lip contours, and vocal tract area function displayed to the left.