I was at a startup from 93-99 until it got sold. One of the things the chip did was music synthesis from wave tables. The traditional way of doing this is to bu

I was at a startup from 93-99 until it got sold. One of the things the chip did was music synthesis from wave tables. The traditional way of doing this is to build a polyphase filter to interpolate between the sample values. Instead, we did the work up-front of converting the wave tables to polynomials. During playback, there was then no need for the polyphase filter -- just evaluate the polynomial piece at whatever fractional phase you care to. As a result, it could synthesize 100+ voices and still have DSP left over. EDIT: I forgot to say, the guy who did that work was Avery Wang, who later went on to invent the algorithm used by Shazam to identify songs.