Overview
Explore the potential of Joint Energy-based Models (JEM) for speech processing in this 12-minute conference talk by Martin Sustek from the Center for Language & Speech Processing at Johns Hopkins University. Gain insights into the advantages of generative models and understand how Energy-based Models fit into this framework. Discover the unique benefits of JEM, which combine the strengths of both generative and discriminative models. Learn about a proposed extension of JEM trained on joint distributions of inputs and phonemes, and its potential applications for multiple tasks without explicit training. Delve into cutting-edge research based on recent work in the field of speech processing and machine learning.
Syllabus
Joint Energy-based Models for Speech -- Martin Sustek (JHU)
Taught by
Center for Language & Speech Processing(CLSP), JHU