ALPHADIGIT RECOGNITION



Utterance Types
  • Isolated Alphadigits => Word Error Rates (WER) below 5%
  • Continuous Alphadigit Utterances => WER between 10% and 20%

  • Approaches
  • HMM-based systems yield best performance
  • Neural Networks
  • Pattern Matching

  • Primary Unit of Recognition
  • Word models - Typically used for isolated Alphadigit recognition
  • Triphone models - Prevalent in continuous Alphadigit recognition
  • Syllable models - Are they applicable??