Abstract: A speech recognition apparatus makes a preliminary selection of a number of candidate words from a vocabulary of words, one of which candidate words is most likely the spoken word to be recognized. For the preliminary selection, each candiate word is divided into first and second portions. For each portion of a word, there are stored probabilites of producing each label of a label alphabet during the utterance of that portion of the word. The speech to be recognized is also divided into first and second portions. A label string representing the speech to be recognized is generated, such that labels occur during the first or the second portion of the speech (or during a transition between the first and second portions. To determine the likelihood that the spoken word represents a word from the vocabulary, each label occurring during the first portion is assigned its "first portion" probability. Each label occurring during the second portion is assigned its "second portion" probability.

Abstract: Speaker adaptation which enables a person to use a Hidden Markov model type recognizer previously trained by another person or persons. During initial training, parameters of Markov models are calculated iteratively by, for example, using the Forward-Backward algorithm. Adapting the recognizer to a new speaker involves (a) storing and utilizing intermediate results or probabilistic frequencies of a last iteration of training parameters, and (b) calculating new parameters by computing a weighted sum of the probabilistic frequencies stored during training and frequencies obtained from adaptation data derived from known utterances of words made by the new speaker.