| 摘要 |
A method of vector classification is disclosed where a test vector is compared with a set of model vectors. Each model vector corresponds to an ideal vector corrupted by different levels of flat spectrum noise. A full band corruption is effectively converted to a series of partial corruptions, and a score is calculated for each model vector based on the matched components. The most severely mismatched components are ignored. The vector classification can take the form of a speech recognition system, or an image recogniser. |
