In order to improve naturalness in speech synthesis, members of the Audio Lab are working on digital waveguide synthesis using tube models that are based on magnetic resonance images of natural speech sounds.
These ‘virtual’ 2-D and 3-D representations can be manipulated in shape to create different dynamic speech sounds such as diphthongs (words like ‘ear’ and ‘eye’).
Physical 3-D prints of these tracts are also being used to test hypotheses relating to the acoustic contributions of different parts of the vocal tract.
These ‘virtual’ 2-D and 3-D representations can be manipulated in shape to create different dynamic speech sounds such as diphthongs (words like ‘ear’ and ‘eye’).
Physical 3-D prints of these tracts are also being used to test hypotheses relating to the acoustic contributions of different parts of the vocal tract.