Early Detection of Continuous and Partial Audio Events Using CNN

Abstract

Sound event detection is an extension of the static auditory classification task into continuous environments, where performance depends jointly upon the detection of overlapping events and their correct classification. Several approaches have been published to date which either develop novel classifiers or employ well-trained static classifiers with a detection front-end. This paper takes the latter approach, by combining a proven CNN classifier acting on spectrogram image features, with time-frequency shaped energy detection that identifies seed regions within the spectrogram that are characteristic of auditory energy events. Furthermore, the shape detector is optimised to allow early detection of events as they are developing. Since some sound events naturally have longer durations than others, waiting until completion of entire events before classification may not be practical in a deployed system. The early detection capability of the system is thus evaluated for the classification of partial events. Performance for continuous event detection is shown to be good, with accuracy being maintained well when detecting partial events.

Robust Deep Learning Frameworks For Acoustic Scene and Respiratory Sound Classification

“…Finally, while the ability to perform early detection (i.e. low latency classification) is very useful for real-world applications such as human-robot interaction, noise reduction during calls or in acoustic security systems, very few publications explore or even mention this [26,27,28]. These several factors motivate this research to develop an ASC system that targets the three most important issues mentioned above while providing a comprehensive analysis on the ability of early detection and exploration of the main factors involved in ensuring good performance.…”

“…The ability of early detecting recording environments. Although early detection of recording environments promisingly opens a wide range of applications as introduced in Section 1.1.1, a few of research [28,27] mentioned and not many experiments have been conducted. This contribution shows an analysis of early detecting recording environments by using the novel Decoder-Encoder framework recently mentioned.…”

“…Inspired by some recent research considering the ability of systems to recognise a sound class early (or using partial data) [27,28], this ability for the different spectrogram types is also evaluated. accuracy for DCASE 2018 Task 1B for all devices and for devices B+C, respectively.…”

CNN-MoE Based Framework for Classification of Respiratory Anomalies and Lung Disease Detection

“…Deep learning techniques have achieved strong and robust detection performance for general sound classification [17], [18]. Feature extraction in state-of-the-art deep learning based systems typically involves generating twodimensional time-frequency spectrograms that are able to capture both fine grained temporal and spectral information as well as present a much wider time context than single frame analysis.…”

Unifying Isolated and Overlapping Audio Event Detection with Multi-label Multi-task Convolutional Recurrent Neural Networks

“…Isolated and overlapping AED literature often appear to derive from two separate methodological streams. For the former, there is a large body of work covering different perspectives: noise robustness [10,7,11], multichannel and multimodal fusion [12,13,14], weak labelling [15,16], early event detection [17,18,19], event detection under scarcity scenarios [20,21], as well as false positive…”