| dc.description.abstract | In this thesis we propose a system with capabilities of simulating guitar sounds with evolutionary methods. Every guitarist has probably at some point had a great desire to be able to recreate the signature sound of their favorite guitar player, that be Mark Knopfler or Kirk Hammett. However, this is not a trivial task to perform manually, and can prove to be both time consuming as well as expensive, considering the wide range of existing guitar effects.
We implement a somewhat simple prototype with a relatively small number of guitar effects. Considering that the perceived sound of the guitar is affected by the order chain of the effects as well as the effects in itself, a small number of effects is preferred to maintain a reasonably low complexity.
An Evolutionary Algorithm and a Sound Synthesis module (consisting of several scripts applying the effects), employing the open-source library Csound, cooperate to evolve individuals representing guitar sounds. The fast Fourier transform is employed in a spectral comparison between the evolved candidate sounds and the specified target sound. In this comparison the \textit{frequency domain} of the sounds are analyzed and compared to 'grade' the candidates.
Varying results concerning the evolved sounds are presented, identifying which effects the system successfully handles, and which ones it struggles to deal with. However, the overall results are promising, and exact matches are eventually found in the majority of the different target sound experiments. | |
