Context dependent modeling of phones in continuous speech using decision trees
HLT '91 Proceedings of the workshop on Speech and Natural Language
Some applications of tree-based modelling to speech and language
HLT '89 Proceedings of the workshop on Speech and Natural Language
CMU robust vocabulary-independent speech recognition system
ICASSP '91 Proceedings of the Acoustics, Speech, and Signal Processing, 1991. ICASSP-91., 1991 International Conference
Hi-index | 0.00 |
Any application involving keyword spotting from continuous speech utterances requires that the wordspotter vocabulary be easily reconfigurable, allowing the wordspotting "task" to change frequently with time. This paper investigates techniques for task independent wordspotter training. Keyword sub-word acoustic hidden Markov models are trained from a very large speech corpus formed from subsets of the TIMIT and General English speech corpora. Decision tree based allophone clustering procedures are used to obtain subword units that are both sensitive to contextual variability and trainable from the available speech data. Task independent wordspotting performance is demonstrated on unconstrained conversational speech utterances.