| 摘要 |
A voice recognizer training and operation procedure including noise identification, involves receiving several noise samples from a number of typical noise environments and then computing acoustic features from the noise samples in a training phase, and then with the noise samples a neuronal network of the voice recognition is trained and during operation of the voice recognition a noise-loaded spoken statement is processed by means of the trained neuronal network and by accessing the noise features data-base, where the acoustic features of the noise environment of the speaker are subtracted. An Independent claim is given for a voice recognizer, especially for speaker-independent voice recognition |
