Learning Algorithms for Speech Processing and Information Theory

Learning Algorithms for Speech Processing and Information Theory

Abstract: The application of Deep Learning to signal processing involves the development of new neural architectures that utilize the basic properties of signals, and often integrates classical signal processing methods. In this talk, I will give an overview of our recent efforts in developing state of the art deep learning methods for text-to-speech, voice translation, vocoders, voice adaptation ,and voice separation. I will present a new architecture for speech synthesis, novel types of neural Diffusion models and new capabilities in speech processing such as mimicking human voices and separating large numbers of speakers.

Shifting the talk toward the intersection of deep learning and information theory, I will present our work that develops neural decoders for error-correcting codes. I will show how we address the decoding problems with a novel neural message passing algorithm combined with Graph Neural Networks and Hyper Networks. Moreover, I discuss the application of the proposed neural decoders for cryptography problems such as AES attacks.

Bio: Eliya Nachmani is a researcher at Facebook AI Research (FAIR) and a Ph.D. student at Tel-Aviv University. His research focuses on machine learning for speech processing and Information theory. He completed his MSc in Electrical Engineering at Tel-Aviv University and BSc in Electrical Engineering at the Technion. 

Website: https://sites.google.com/view/eliya-nachmani/home