Generating New Sounds by Vector Arithmetic in the Latent Space of the MelGAN Architecture

In this paper, we investigate the exploitation of the latent space of a MelGAN architecture by the vector arithmetic for the generation of new sounds that may be appealing for musicians, similar to what has already been done in the case of words and images. Specifically, since the MelGAN uses directly the spectrogram as input to its generator, we focus our attention on the linear combination of two or three instrumental sounds. This combination is then fed to the MelGAN generator, and the produced output will be the new sound with innovative sonority. Some simulations, performed over different sounds and different combination coefficients, show the effectiveness of the proposed idea.