| 摘要 |
The present invention provides a system for recognizing speech in which a profile is constructed of a related characteristic of each significant phoneme in a language. A difference profile is generated for each pair of significant phonemes by subtracting the profile of each phoneme from the profile of each other phoneme. Adjacent sections of each difference profile which exceed positive and negative thresholds are identified. The likelihood that an unknown phoneme will be one or the other of a phoneme pair based on the relative areas in the identified sections of the difference profile is computed. An equivalent profile is constructed of a phoneme of an unknown utterance. The most likely phoneme of each phoneme pair is chosen based on the relative areas in the identified sections of the profile of the unknown phonemes. |
