Complexity optimization technique for sound synthesis on digital sound processing architectures

Synthesized sounds are practically used for many applications, such a HMI using a voice dialog, broadcasting of synthesized music and voices. Natural and realistic sounds are required for the applications and it is necessary for some applications, such as two or more voices dialog and music, to produce many sounds at one time. One generating method of realistic sounds is to generate them by simulation of an instrument or a human voice using its sound production model. Appearance of sound processing processors such a DSP (digital signal processor) enabled a practical method to use for only a single sound generation. There is a problem, however, in that the computational complexity of the simulation is large. The authors research a new technique for the solution of the problem. It is a synthesizer optimization technique which gets a sample sound, determines a suitable synthesis for producing the sound, sets up and optimizes parameters and tables, and generates an optimum synthesizer automatically. If a simulated sound is given for input, it can be produced by the synthesizer with smaller complexity than the simulation. The authors categorize the synthesis methods, and each complexity is discussed. The synthesizer optimization technique is constructed from this discussion. This technique is evaluated by comparing the complexity of the generated synthesizers and that of existing synthesizers. The result of the evaluation is that the optimized synthesizer has from 50% to 70% of the complexity of the existing synthesizer on average.