| 摘要 |
<p>PROBLEM TO BE SOLVED: To provide an abnormal conversation detector capable of determining whether a conversation is abnormal or not, on the basis of voice signals recording the conversation of multiple speakers.SOLUTION: An abnormal conversation detector 1 includes: a voice input section 11 which inputs voice signals containing a conversation of multiple speakers; a storage section 14; a feature quantity extracting section 22 which extracts at least two feature quantities expressing a feature of a person's voice in a frame unit having a prescribed time length, with respect to the voice signals, and stores a set of at least the two feature quantities in the storage section 14; a fitting section 23 which fits a distribution of the feature quantities stored in the storage section 14 at a probability distribution of the same number as a number of speakers; and a determination section 24 which determines whether or not the probability distribution of the same number as the number of speakers approximates the distribution of the feature quantities, and in the case that the probability distribution of the same number as the number of speakers does not approximate the distribution of the feature quantities, determines that the conversation is an abnormal conversation.</p> |
