Each of these sentences was produced by combining the speech of two speakers with very different characteristics. Only one of the two speakers has the voice characteristics you perceived. These sentences are examples of our innovative hybrid speech synthesis, in which only those speech segments required to characterize a speaker are actually taken from that speaker's speech. The remaining segments can be taken from an entirely different source, be it another speaker or computer-generated speech. In the example sentences, at least half of the segments were taken from other sources.

Our ability to create these illusions results from our team's extensive experience in knowledge-based speech synthesis and related technologies. Building on this experience, we can predict for any given language which aspects of a speech signal and which types of signal manipulations will be perceptible to the human ear and which will not. Building on a powerful set of perceptual models, we are currently developing improved methods of speech synthesis and speech encoding, and are planning on extending our novel methods into other areas of speech processing.