Difference between Human and Machine in Feeling about Similarity of Melodies

抄録

In this paper, we report the experimental result of examining the difference of feeling of human and machine to the similarity of melody. First of all, original melodies are divided into 4 groups for each work, and fake melodies similar to original ones are generated by Deep Convolutional Generative Adversarial Net (DCGAN). At that time, the discriminator of each GAN is learned so as to be able to evaluate the similarity with the work which is not learned. We ask ten subjects to evaluate impressions for melodies generated GANs, and calculate the similarity between melodies. We compare the similarity evaluation by human and that by machine.