0562138-A1 is referenced by 15 patents.
A method and apparatus for generating a word model to be used in a speech recognizer based on the spelling of the word and one utterance of the word by the user, are described. In a first step a list of all the possible phonetic transcriptions for the new word is produced using phonotactical knowledge of the language. In a second step a score is computed for each transcription of the list, combining the score obtained matching the utterance against the Hidden Markov Models derived from its possible phonetic transcriptions and the a priori probability of the phonetic transcription. The phonetic transcription with the highest score is selected as the correct one.