A parallel processing system for polyphonic singing synthesis

A synthesiser has been designed for polyphonic (multi-voice) singing synthesis. Singing synthesis requires frequent updating of parameters to achieve a natural output sound. These parameters include fundamental frequency, formant frequencies, formant bandwidths, and voice source parameters (such as Open Quotient). The synthesis engine is controlled using a script and a graphical user interface. The hardware consists of several SHARC ADSP-21060 processors, an FPGA and a multi-channel D/A convertor. This paper presents an overview of the synthesiser and the synthesis model employed. The results show a soundfile which has been created using a synthesis-by-analysis approach