by Hiroki Nishino.

We propose a novel implementation technique, speculative digital sound synthesis, as a practical solution for the tradeoff between computational efficiency and sample-rate accurate control in sound synthesis. Our technique first optimistically assumes that there will be no change to the control parameters for sound synthesis and computes by audio vectors at the beginning of a DSP cycle. Then, after the speculation, when any change is made in the same cycle, it recomputes only the neces- sary amount of the output.

As changes to control parameters are normally quite sporadic in most situations, recomputation is rarely performed. Hence, the computational efficiency can be maintained mostly equivalent to the computation by au- dio vectors without any speculation, when no changed is made to the control parameters. Even when any change is made, the overhead can be reduced since the recomputation is only applied to those sound objects that had their control parameters updated, and the output samples in the past are not recomputed.

Thus, our speculative digital sound synthesis technique can provide both sample-rate accurate control in sound synthesis and better performance efficiency by the audio vectors in most practical situations. The tradeoff between these two issues has been a long-standing problem in computer music software design.

Publication:
H. Nishino & A. Cheok, Speculative Digital Sound Synthesis, The 13th Sound and Music Computing Conference (SMC 2016), Hamburg, Germany, 2016 Aug. pp.358- 365.