A Multinomial View of Signal Spectra for Latent-Variable Analyses

Explore a novel mathematical framework for analyzing signal spectra through a multinomial perspective in this research lecture delivered at Johns Hopkins University's Center for Language & Speech Processing. Learn how to apply latent-variable analysis techniques to signal processing problems by treating spectral data as multinomial distributions. Discover the theoretical foundations and practical applications of this approach for understanding hidden structures in audio and speech signals. Examine how this multinomial view can enhance traditional spectral analysis methods and provide new insights into signal decomposition and feature extraction. Gain exposure to advanced mathematical concepts that bridge probability theory and signal processing, presented by a researcher from MERL Research Lab with expertise in machine learning applications to audio processing.