The genetic algorithm for timbre mimicking imitates a target sound, exclusively using grains from a source sound.
The intermediate generations of the algorithm move through a timbre space best expressed by the uncanny valley, a concept borrowed from animation and robotics. It explains that the more human a manufactured object gets, the more we like it, except when it imitates humans closely but not perfectly. Then, it's considered creepy instead.
Here's a sample run of the algorithm, laid out in the uncanny valley:
chopped up, mimicking
The Uncanny Valley
Preference vs. Fitness
Fitness
Click a data point to hear the corresponding algorithm output.
As the timbre mimicking algorithm starts to approach the target, the output becomes more pleasing. However, when the mimicked timbre is close but still not quite right, it's unsettling instead.
In the language of genetic algorithms: as fitness increases, the organisms are more pleasing (generation 1-40). However, beyond a certain fitness, the organisms are perceived as unsettling because they are close enough to the target to be perceived as an imitation, but not close enough to fool us (generation 1000).
By timbre morphing the final output of the algorithm (generation 1000) with the target sound, we exit the uncanny valley as the sound becomes more pleasing again.