Deep CNN Framework for Audio Event Recognition using Weakly Labeled Web Data

TL;DR

This paper introduces a deep CNN framework capable of learning audio event recognition from weakly labeled web data, effectively handling variable-length recordings and localizing events without precise annotations.

Contribution

It presents a novel deep CNN approach that trains on weak labels and localizes events during inference, outperforming models trained on strongly labeled data.

Findings

Outperforms models trained on strongly labeled web data

Efficiently analyzes variable-length recordings

Capable of localizing events without explicit time annotations

Abstract

The development of audio event recognition systems require labeled training data, which are generally hard to obtain. One promising source of recordings of audio events is the large amount of multimedia data on the web. In particular, if the audio content analysis must itself be performed on web audio, it is important to train the recognizers themselves from such data. Training from these web data, however, poses several challenges, the most important being the availability of labels: labels, if any, that may be obtained for the data are generally weak, and not of the kind conventionally required for training detectors or classifiers. We propose that learning algorithms that can exploit weak labels offer an effective method to learn from web data. We then propose a robust and efficient deep convolutional neural network (CNN) based framework to learn audio event recognizers from weakly labeled data. The proposed method can train from and analyze recordings of variable length in an efficient manner and outperforms a network trained with strongly labeled web data by a considerable margin. Moreover, even though we learn from weakly labeled data, where event time stamps within the recording are not available during training, our proposed framework is able to localize events during the inference stage.