whole-word based recognition
| A recognition technique which is based on comparing input utterance with the Markov models associated with each word in the active vocabulary. The Markov models are trained from several spoken examples of these words or expressions (several hundred examples in the case of speaker-independent recognition).
| |
| word spotting
| A recognition system which is only interested in certain parts of the utterances: those parts which contain application "trigger" words. In these systems, the utterances are modeled as out of vocabulary sequences (i.e. sequences which are to be rejected) and valid words (not to be rejected). The term multi-word spotting is used when the system can recognize more than one trigger word in the same utterance (e.g. multi-field information).
| |
|
