State-of-the-art automatic speech recognition systems are based on statistical approaches, and use hidden Markov models (HMMs) for acoustic modelling. These acoustic models are trained from a large amount of speech data usually collected from a number of speakers under different acoustic environments. The training data contain both the desired variabilities for speech recognition as well as noise and other unwanted variabilities from speakers and the environment. Moreover, there is usually a mismatch in training and testing conditions. Therefore, adaptation and adaptive training of acoustic models plays an important role in speech recognition systems. In this talk, first, maximum-likelihood based adaptation and adaptive training will be briefly reviewed, and then discriminative and Bayesian approaches to adaptation will be discussed.
C. K. Raut received PhD in Information Engineering from Cambridge University. He also holds Master in Information Science and Technology, Bachelor in Electronics Engineering, and Diploma in Electrical Engineering degrees. He has an experience of working as a scientist in the area of speech and language processing at Raytheon BBN Technologies, Massachusetts, USA. In the past, he also served as a lecturer cum head of department at the department of computer engineering at a private college. He was involved in the US Department of Defense/DARPA funded Global Autonomous Language Exploitation (GALE) and Robust Automatic Transcription of Speech (RATS) projects. His research area includes audio signal processing, robust speech recognition under noise and reverberation, acoustic modelling, adaptive training, and model adaptation. He is also a recipient of the NEA Young Engineer Award.