HMM LIMITATIONS: CONDITIONAL INDEPENDENCE

Recall our basic acoustic model topology:
It can be argued that HMMs do not provide a realistic model for the temporal structure of speech: What can we do to overcome these deficiencies?