Research in Continuous Speech Recognition.
Annual rept. 1 Oct 82-30 Sep 83,
BOLT BERANEK AND NEWMAN INC CAMBRIDGE MA
Pagination or Media Count:
This annual report describes the work performed during the past year in an ongoing effort to design and implement a system that performs phonetic recognition of continuous speech. The general approach used it to develop a Hidden Markov Model HMM of speech parameter movements, which can be used to distinguish among the different phonemes. The resulting phoneme models incorporate the contextural effects of neighboring phonemes. One main aspect of this research is to incorporate both spectral parameters and acoustic-phonetic features into the HMM formalism. Author