| 摘要 |
A speech recognition system is provided having a training mode in which users can add new vocabulary words to the recognition system. The training system includes a system which compares a new model with each of the existing models to ensure that a user is warned if, during the training mode, a new word is added which is acoustically similar, and hence confusable, with an existing vocabulary word. This model comparison system uses a measure of the variation between renditions of training utterances for the new word and uses this in its calculations. <IMAGE> |
