Applying a hybrid intonation model to a seamless speech synthesizer

We present a speech synthesizer to seamlessly concatenate recorded and synthetic phrases to produce natural sounding and highly expressive speech. Not only the acoustic units, but also the F0 contours are seamlessly concatenated together from recorded and synthetic phrases. When mixed with recorded phrases, the F0 contours of synthetic phrases are generated adaptively relative to the actual surrounding F0 shapes of the recorded phrases. Although the intonation generation scheme was originally developed for unlimited speech synthesis, it is quite naturally extended to a hybrid intonation generation.