Verfahren zur Spracherkennung

摘要

The features generated for a new speaker (training sample) during the training of a speech recognition system are compared with a recognition sample for speech recognition generated on the basis of a speech signal recorded at a later date. These features are optionally linked with signals that are analog to the noise that may occur during recording of the speech signal and that may falsify the speech signal. The inventive method is further characterized in that the training sample is adapted to the recognition sample that is likewise generated by linking the speech signal with signals that are analog to the noise.