Implementation of Viterbi decoding for the number recognition
(Machine Learning, Korea University)
- Transition probabilty was pre-trained by Baum-Welch algorithm
- Observation probability is defined by Gaussian Mixture Model
- Transition probability within words was defined by unigram and bigram probability.
Results
99.88% accuracy in words and 96.38% accuracy in sentences.
