Hands-on Guide To GANSynth: An Adversarial Neural Audio Synthesis Technique

 GANSynth is a state-of-the-art method for synthesizing high-fidelity and locally coherent audio using Generative Adversarial Networks (GANs). Hence the name GANSynth (GAN used for audio Synthesis). It was introduced by Jesse Engel, Kumar Krishna Agrawal, Shuo Chen, Ishaan Gulrajani, Chris Donahue and Adam Roberts - researchers at the Google AI in 2019 (research paper). Autoregressive models like WaveNets generate audio sequentially. On the contrary, GANSynth creates the whole sequence in parallel, synthesizing audio much faster on GPU runtime than real-time synthesis. It generates the entire audio clip from a single latent vector, allowing for easier release of global features like pitch and timbre (tone quality). It uses progressive GAN architecture. It eliminates the drawback of