churia / nlp-discrete_hmm

Discrete HMM for Speech Recogition

In acoustic model,

• each word consists of syllables
• each syllable consists of phonemes
• each phoneme consists of some (hypothetical) states.

Each phoneme can be described by a HMM (acoustic model). Each time frame, with an observance (MFCC vector) mapped to a state. Hence, there are state transition probabilities a_ij and observation distribution b_j(o_t) in each phoneme acoustic model.

Usually in speech recognition we restrict the HMM to be a left-to-right model, and the observation distribution are assumed to be a continuous Gaussian mixture model. Here we use characters to simulate the observance.

training HMM

• Algorithm: Baum-Welch
• input:
  • observed sequences (seq_model_01~05.txt)
  • initial state & transition probabilities (model_init.txt)
  • number of iteration
• output:
  • trained models for 5 sequences

testing HMM

• Algorithm: Viterbi
• input:
  • trained model
  • testing sequences
  • modellist.txt
• output
  • best labels (with the higest probability among the five models)

Experiment

see the document