A Joint Detection-Classification Model for Audio Tagging of Weakly Labelled Data
This addresses the challenge of accurately tagging audio clips without event-level labels, which is incremental as it builds on prior bag-of-frames models by adding detection capabilities.
The paper tackles the problem of audio tagging with weakly labelled data by proposing a joint detection-classification model that attends to informative sounds and ignores uninformative ones, reducing the equal error rate from 19.0% to 16.9% on the CHiME Home dataset.
Audio tagging aims to assign one or several tags to an audio clip. Most of the datasets are weakly labelled, which means only the tags of the clip are known, without knowing the occurrence time of the tags. The labeling of an audio clip is often based on the audio events in the clip and no event level label is provided to the user. Previous works have used the bag of frames model assume the tags occur all the time, which is not the case in practice. We propose a joint detection-classification (JDC) model to detect and classify the audio clip simultaneously. The JDC model has the ability to attend to informative and ignore uninformative sounds. Then only informative regions are used for classification. Experimental results on the "CHiME Home" dataset show that the JDC model reduces the equal error rate (EER) from 19.0% to 16.9%. More interestingly, the audio event detector is trained successfully without needing the event level label.