| 摘要 |
Usually in a speaker adaptive system, every time a change in speaker occurs, he/she has to chose which of the available model sets to use. E.g. the SI model set, if it is the first time he/she is using the system or if a model set already adapted to him, if he/she used it before. If adapted model sets are not stored at all, the adaptation process starts over and over again using the SI models, if the same speaker uses the system repeatedly. According to the invention a change in speaker is be detected automatically. Furtheron, the system identifies the speaker and if he/she had used the system before and a speaker adapted model to him/her is already available. If this is the case, this model set will be taken for further recognition and adaptation. <IMAGE> |
