用遗传算法对乐音合成技术的改进

Improvement of the Synthsis Technique of Musical Instrument Tones Using Genetic Algorithms

乐音合成方法有采样、频率调制、波表合成和离散求和合成等。频率调制控制参数少，容易实现，但很容易被辨出是电子音乐；波表合成可以很逼真，但需要存储波表；离散求和合成不需要波表的存储，但若参数选取不当，模块数较少，它的相对误差比波表合成大。遗传算法是优化算法，对问题的依赖性小，可以求得全局最优解。本文提出采用遗传算法寻找并优化离散求和合成参数，实现乐音的合成，以改进乐音舍成的效果。使用二胡的乐音进行实验，并与其他方法作了比较。

Synthesis techniques are sampling, frequency modulation, wavetable synthesis and discrete summation synthesis. Frequency modulation has less number of parameters and can be easily implemented, but it is easy to be identified as electronic music. The result of wavetable synthesis is very natural and real. Wavetable synthesis requires for storage of wavetables whereas discrete summation synthesis does not. If the parameters are chosen unsuitable and there are fewer modules, the relative errors of discrete summation matching tones are more than that of wavetable matching tones. Genetic algorithms are opti-mum algorithms, which depend little on questions and are able to find global optimum, In order to imple-ment the synthesis of musical instrument tones and modify its results, this paper proposes to use genetic algorithms to find and optimize the parameters of discrete summation synthesis. Finally, the synthesis result of erhu instrument tones is given out in comparison with other methods.