Patent US 6,236,966

A method and device for producing audio control parameters from symbolic representations of desired sounds includes presenting symbols to multiple input windows of a learning machine, where the multiple input windows comprise a lowest window, a higher window, and possibly additional higher windows. The symbols presented to the lowest window represent audio information having a low level of abstraction (e.g., phonemes), and the symbols presented to the higher window represent audio information having a higher level of abstraction (e.g., words or phrases). The learning machine generates parameter contours and temporal scaling parameters from the symbols presented to the multiple input windows. The parameter contours are then temporally scaled in accordance with the temporal scaling parameters to produce the audio control parameters. The techniques can be used for text-to-speech, for music synthesis, and numerous other applications.