To complete the model, We used Stem-ML step_to tags to implement the phrase curve, and one phrase tag was placed at the end of the utterance. Four other Stem-ML parameters control overall properties: adroop (controls the maximum time interval over which one stress tag can affect the pitch), add (controls how overlapping stess tags add together), smooth (how fast the muscles can respond), and base (the speaker's normal resting pitch). These are all free parameters in the fit.

Greg Kochanski, Chilin Shih 2002-08-03