Automatic Speech Recognition ASR

1
Automatic Speech Recognition (ASR)

• Issues
• Architecture
• Main Focus Implementing a basic system
• Hidden Markov Models (HMMs)
• Tying things together
• Reading
• Rabiner Huang
• Jurafsky Martin
• Owen

2
ASR Issues

• Speaker dependent/independent
• Vocabulary size
• Isolated word vs. Continuous Speech
• Language modelling constraints
• Level of ambiguity in vocabulary
• Environment e.g. noise considerations

3
Acoustic Phonetic Approach
4
Acoustic Phonetic Approach

• Requires knowledge of acoustic properties
• Ad hoc choice of features
• No automatic way of optimally tuning ethods,
  decisions, etc.
• Not used nowadays for practical applications

5
Statistical Pattern Recognition
6
Statistical Pattern Recognition

• Performance related to training data
• Reference patterns sensitive to environment
• No speech-specific knowledge required
• Computation can be heavy
• Search strategies important
• Same algorithms can be applied to variable length
  units
• HMMs most widely used
• Dynamic Time Warping
• Template Matching voice dialling