Скачать или смотреть Speech Processing - L23 - Time-Domain Methods - The Dynamic Time Warping (DTW)

Dr. Agha Ali Raza (https://aghaaliraza.com/) delivered this Speech Processing lecture series at the Lahore University of Management Sciences (LUMS) in Spring 2022. This course offers a theoretical and practical understanding of how human speech is processed using computers. Speech Processing lies at the intersection of articulatory and acoustic phonetics, digital processing of speech signals, and Machine Learning. Knowledge of these domains is essential to developing a thorough understanding of the rapidly growing fields of speech recognition (speech-to-text), speech synthesis (text-to-speech), spoken dialog systems, and chatbots (such as Siri, Alexa, and Cortana). Students will learn about the processes underlying human speech production, perception, acoustics, speech analysis, and synthesis techniques.

This offering focuses on Urdu as the main example language. However, all major concepts are situated using examples from multiple languages, including English, Urdu, Hindi, Arabic, and Punjabi.

The course was designed to be accessible with a low bar for entry yet rigorous in terms of the mathematical foundations and hands-on programming experience. The language of the lectures is a mix of Urdu/Hindi and English to allow for maximum dissemination among interested audiences comfortable with these languages. For more information about the course, including textbooks, readings, assignments, homework problems, and acknowledgments, please visit https://www.c-salt.org/courses/.

Time-Domain Methods – Dynamic Time Warping
• Euclidean Distance (L2 Distance)
• Dynamic Time Warping
• Procedure
• The Distance Matrix
• Alignment
• The Algorithm
• Additional Constraints
o Monotonicity
o Continuity
o Boundary
o Window
o Slope
• Segmental DTW