Video Recordings

The source signal in conversational speech carries syntactic, semantic, and emotional information. While the source/filter theory of speech synthesis has a long history, but the technology used has failed to separate the source from the filter in a convincing way. New methods in speech morphing promise a cleaner separation between the source and filter using STFT techniques.  I will explain the voice morphing system, and demonstrate how to derive the source signal from speech. Details of the source signal itself can be examined for clues to the semantic representation of the resulting utterances.