New Methods to Capture and Exploit Multiscale Speech Dynamics - From Mathematical Foundations to Practical Applications

Explore advanced mathematical approaches to analyzing speech patterns across multiple time scales in this lecture from Harvard University's Statistics and Information Sciences Laboratory. Learn about innovative computational methods for capturing the complex temporal dynamics inherent in human speech signals, from short-term acoustic features to longer-term linguistic structures. Discover how multiscale analysis techniques can reveal hidden patterns in speech data and understand their applications in speech processing, recognition, and synthesis. Examine the mathematical foundations underlying these new methodologies and their potential to advance our understanding of speech production and perception mechanisms.