Expressive and Conversational Speech Synthesis

For virtual characters to be believable in spoken conversation, they need to sound like real people in conversation. For limited domain characters this can be achieved by having humans record the lines the characters say. For characters with a bigger or more open, flexible domain, this is not practical, and we must rely on speech synthesis. In this effort, we partner with external speech synthesis experts to make synthesizers more suitable for conversational speech, including variations in emotional expression, and disfluencies, such as filled pauses, fillers, etc.