| CPC G06N 3/088 (2013.01) [G10L 25/03 (2013.01); G10L 25/51 (2013.01)] | 19 Claims |
|
1. A system, comprising:
at least one processor; and
a memory, storing program instructions that when executed by the at least one processor, cause the at least one processor to:
receive audio data via an interface for an anomaly detection system;
divide, by the anomaly detection system, the audio data into different frames;
apply, by the anomaly detection system, an autoencoder network model trained without supervision to output an estimated joint probability distribution of normality for a group of the different frames, wherein one or more connections in the autoencoder network model are masked to enforce a dependency ordering between individual ones of the group of frames to estimate the joint probability distribution of normality, and wherein the dependency ordering estimates the joint probability distribution for one frame of the group of frames according to one or more frames of the group of frames that occur after the one frame;
determine, by the anomaly detection system, an anomaly score based, at least in part, on the estimated joint probability distribution of normality for the group;
detect, by the anomaly detection system, an anomaly in the audio data according to the determined anomaly score; and
send, by the anomaly detection system, a result via the interface indicating the detected anomaly in the audio data.
|