| 摘要 |
<P>PROBLEM TO BE SOLVED: To determine whether an observed sound is a target sound to be recognized or not without omission. <P>SOLUTION: Likelihoods to a first recognition model based on objective sounds and a second recognition model based on non-objective sounds with respect to a feature quantity extracted from observed sounds by using an HMM and a recognition result showing which of objective sounds and non-objective sounds the observed sounds are, are obtained by a recognition unit, and the likelihoods corresponding to recognition models indicated by the recognition result are used to calculate a degree of reliability of the recognition result in a reliability degree calculation unit. The recognition result and the degree of reliability are used to calculate, in an accuracy rate calculation unit, an accuracy rate of the recognition result on the assumption that the observed sounds are objective sounds and on the assumption that the observed sounds are non-objective sounds. On the basis of the recognition result and the accuracy rate, an alarm output processing unit outputs information indicating that the observed sounds are unknown sounds belonging to neither objective sounds nor non-objective sounds, in the case that the recognition result indicates that the observed sounds are objective sounds and the accuracy rate is equal to or lower than a first threshold and in the case that the recognition result indicates that the observed sounds are non-objective sounds and the accuracy rate is equal to or lower than a second threshold. <P>COPYRIGHT: (C)2012,JPO&INPIT |
