Automatic speech processing

access password

16.ASP+lec9$11

In this part of lecture, we looked at the formulation of ASR problem as selecting a best matching word hypothesis from a set of word hypotheses. We then focused on the problem of how to model set of word hypotheses. For that, the lecture dwelt into discrete Markov model and the key theoretical aspects. We then saw how DMM can be applied to model word sequences, i.e. language modeling. Finally, we briefly discussed how neural language models or large language models (LLMs) are extension of this problem.