GAN-based models for audio generation