Abnormal sound detection method for restraining self-supervised classification by utilizing metadata hierarchical information

• Main Authors: GUAN JIAN, XIAO FEIYANG, TIAN JIANTONG, LAN HAIYAN, WEI YUMING
• Format: Patent
• Language: Chinese; English
• Published: 17.11.2023
• Subjects: ACOUSTICS; CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; ELECTRIC DIGITAL DATA PROCESSING; MUSICAL INSTRUMENTS; PHYSICS; SPEECH ANALYSIS OR SYNTHESIS; SPEECH OR AUDIO CODING OR DECODING; SPEECH OR VOICE PROCESSING; SPEECH RECOGNITION

The invention belongs to the technical field of abnormal sound detection, and particularly relates to an abnormal sound detection method for restraining self-supervised classification by utilizing metadata hierarchical information. According to the method, a metadata hierarchical information structure constraint neural network is used for learning low-dimensional features and high-dimensional features of a training audio, metadata accompanying an audio file is fully utilized, and the influence of metadata attributes on acoustic features is mined, so that the neural network can learn changes caused by domain offset on the audio features, and the accuracy of the audio feature learning is improved. And thus, the performance of the industrial abnormal sound detection system under the domain offset condition is improved. Meanwhile, the invention provides an abnormal score calculation method taking the attribute group as the center, and the method is used for evaluating the abnormal score of the test sample under t