Synthesis of Speech Sounds from a Multi-mass Model of the Lungs, Vocal Tract, and Glottis

Like some existing vocal-tract models, this synthesizer views the vocal apparatus as a sequence of straight tubes. The walls of all these tubes, however, are modelled as coupled mass-spring systems, which in most models is privileged to the vocal cords. This, together with the ability to continuously vary the lengths of the tubes, leads to a principled treatment of the problems of source-filter interaction, which makes our model especially suitable for the generation of speech signals that include consonants. The model is restricted to the last step in articulatory synthesis: from a list of time­ target pairs for every articulator and a table with speaker characteristics, find the resulting acoustic signal. Among the examples that we show are various brands of voicing contrasts and their articulatory correlates.