A particular Google blog post reads that this experiment makes use of a neural network that is trained on the different voices of tenor, bass, soprano, and even mezzo-soprano in order to generate ...