From Bases to Exemplars, and From Separation to Understanding

Explore the evolution of audio signal processing techniques in this seminar lecture that traces the progression from basis decomposition methods to exemplar-based approaches, and from traditional source separation to deeper audio understanding. Learn about advanced computational methods for analyzing and processing audio signals, including non-negative matrix factorization, sparse coding, and exemplar-based techniques. Discover how these mathematical frameworks can be applied to solve complex problems in audio source separation, music information retrieval, and automatic speech recognition. Examine the theoretical foundations behind basis learning algorithms and understand how exemplar-based methods provide more flexible and powerful alternatives for audio analysis. Gain insights into the shift from purely separating audio sources to developing systems that can understand and interpret audio content at a semantic level. The presentation covers both the mathematical underpinnings of these techniques and their practical applications in real-world audio processing scenarios.