Please enable JavaScript.

Coggle requires JavaScript to display documents.

CSI (Speech Recognition (Real World Issues (Adaptation (needed when there…

- - - - markov model : only two elements, and transition depends only on current state
      - HMM : in addition there is initial transition probability , emission probabilty and feature space
  - - - Statistical
    - - Rule of Language
  - - - needed when there is a mismatch between training and testing data
      - used when we don't want to repeat the training for new variables (example: American English and British English )
- - - - Training
      - Testing
    - - Supervised (accurate but costly)
      - Un-Supervised (not accurate but cheap)
  - - - Learning Paradigm
- - - - why is it challenging
      - advantages
      - definition
      - when is it important
      - difference between multimodal and multimedia
  - - - what is lip reading
      - how is it multimodal
      - why we do it
      - how it works (steps for audio and visual, then fusion)
      - early fusion and late fusion : the philosophy (more general)
        
        early fusion : signal at feature level
        
        late fusion : info at semantic level
    - - helpful in noisy environment
      - used only for few words