I think it's supposed to be some kind of latent vector space where the coordinate maps to the sounds' timbre. So in theory, you can move seamlessly between soun

I think it's supposed to be some kind of latent vector space where the coordinate maps to the sounds' timbre. So in theory, you can move seamlessly between sounds. But I seem to only be able to snap to nearest sound. Edit: they use t-SNE for the embedding so that is why it is discrete. But if they used something like an autoencoder, then you could have a smooth continuum and interpolate between say a dog bark and a billiards clack.