| 摘要 |
A set of features is extracted from an input portion of speech provided by a speaker. A first scoring means 4 scores the set of features with a first stored model of mixture components derived from sets of features extracted from input portions of speech provided by a plurality of speakers. A second scoring means 12 scores the set of features with a second stored model of mixture components derived from sets of features extracted from input portions of speech provided by the speaker to be identified. The results are compared, 16, to determine whether the input portion of speech did originate from that particular speaker. The first scoring means 4 scores the set of features with only part of the first stored model most likely to provide a good match to the set of features provided. |
