Cooperative Learning Of Deep Generative Models With Application In Sound Synthesis