| 摘要 |
A speaker-dependent word recognition system (100) operable in a severe acoustical environment compares linear predictive parameter signals representative of unknown utterances spoken by an operator with similar parameter signals representative of a predetermined library of reference utterances, and accepts or rejects identification of the unknown utterances as one of the reference utterances on the basis of the comparison. Sets of linear prediction coefficient parameter signals representative of the reference utterances are stored in a reference store (104). Signal conditioning circuitry (114, 124, 128, 134, 138, 140) including a controllable gain amplifier (128) converts analog signals representative of the unknown utterance into digital sample signals. The linear prediction coefficient parameter signals include sets of partial correlation (PARCOR) coefficient signals derived from the digital sample signals, and distance signals are generated which represent the dissimilarity between the derived PARCOR signals and the reference utterance PARCOR signals. An averaged distance signal representative of the relationship between the derived PARCOR signals and the PARCOR signals of the most similar reference utterance is compared with a predetermined threshold signal. If the level of the averaged distance signal is below the threshold signal level, word command signals are generated and display signals indicate the identified reference word to the operator. |
