Learning a female utterance with a male vocal tract
Top left: English word "aware" spoken by a female speaker in American accent
Top right: Synthetic "aware" generated by VocalTractLab with articulatory parameters of a male speaker learned directly from the acoustics (MFCC) of the female utterance, with no speaker normalization
Upper middle: Animation of gradual spectrographic improvement as articulatory parameters were repeatedly adjusted
Lower middle: Animated vocal tract movements of the synthetic utterance
Bottom: Continuous movements of individual articulators
Original, Female
Synthesis, Male
Spectrogram
Sound
Video: Learning articulatory targets of a virtual male speaker from female speaker data
Video: Male vocal tract movements synthesized from learned articulatory targets (slow motion)
Articulatory gestures synthesized from learned articulatory targets