Low-dimensional real-time concatenative speech synthesizer

摘要

A method of providing real-time speech synthesis based on user input includes presenting a graphical user interface having a low-dimensional representation of a multi-dimensional phoneme space, a first dimension representing degree of vocal tract constriction and voicing, a second dimension representing location in a vocal tract. One example employs a disk-shaped layout. User input is received via the interface and translated into a sequence of phonemes that are rendered on an audio output device. Additionally, a synthesis method includes maintaining a library of prerecorded samples of diphones organized into diphone groups, continually receiving a time-stamped sequence of phonemes to be synthesized, and selecting a sequence of diphone groups with their time stamps. A best diphone within each group is identified and placed into a production buffer from which diphones are rendered according to their time stamps.