Introduction to Automatic Speech Recognition

Jean-Luc Rouas from LaBRI presents an in-depth overview of automatic speech recognition (ASR). The presentation begins with a historical perspective, covering early systems like “Audrey” and “Shoebox: and progresses to modern neural techniques. It explains key concepts such as Hidden Markov Models (HMM), Dynamic Time Warping, and end-to-end neural architectures. The talk emphasizes the evolution from isolated word recognition to continuous speech recognition and the challenges posed by speech signal variability. It also introduces model compression techniques and their application within the FVLLMONTI project aimed at optimizing ASR models for low-resource conditions.