Really interesting ! Regarding the sound synthesis application, I was thinking that maybe using some FFT analysis could work as a fitness function (avoiding too high frequencies, having well defined harmonics, etc.)
@@ValerioVelardoTheSoundofAI Yeah totally ! I think there's also a measure for "roughness" of sound that comes from psychoacoustics. That could be usefull as well :)
Say, for example, that you wanted to create a 1 bar melody (in 4/4) and you spliced two parent genes together based on note duration, isn't there a chance that the result would be more than 1 bar long? Would you use a fitness function to avoid such cases? Thanks
You could use the fitness function to punish irregular durations, or you could use an encoding that ensures you'll always end up with 1 bar. E.g., having a chromosome with 16 genes, where each gene is a 16th note.
Can't wait to implement it with you
Really interesting ! Regarding the sound synthesis application, I was thinking that maybe using some FFT analysis could work as a fitness function (avoiding too high frequencies, having well defined harmonics, etc.)
That could be really cool to try out!
@@ValerioVelardoTheSoundofAI Yeah totally ! I think there's also a measure for "roughness" of sound that comes from psychoacoustics. That could be usefull as well :)
Say, for example, that you wanted to create a 1 bar melody (in 4/4) and you spliced two parent genes together based on note duration, isn't there a chance that the result would be more than 1 bar long? Would you use a fitness function to avoid such cases? Thanks
You could use the fitness function to punish irregular durations, or you could use an encoding that ensures you'll always end up with 1 bar. E.g., having a chromosome with 16 genes, where each gene is a 16th note.
Great, thanks!@@ValerioVelardoTheSoundofAI