This course on Automatic Speech Recognition covers topics such as pronunciation modeling, acoustic modeling, and language modeling. Students will learn about voice dialing systems, dynamic time warping, matching templates, training an acoustic model, and estimating the cost of a sequence of words in a language. The teaching method involves discussing various ASR components and their reliability, as well as measuring ASR success. This course is intended for individuals interested in multilingual natural language processing and automatic speech recognition.
Overview
Syllabus
Automatic Speech Recognition
Voice Dialing System
Matching in Frequency Domain
Dynamic Time Warping
DTW algorithm
Matching Templates
DTW issues
More reliable matching
More reliable distances
Extending template model
Training an acoustic model
Language Model Estimate cost of sequence of words in the language • Need appropriate training data
Pronunciation Model
Measuring ASR Success
How good is good?
ASR Discussion Point
Taught by
Graham Neubig
