Detection and Classification of Acoustic Scenes and Events

Abstract

International audience—For intelligent systems to make best use of the audio modality, it is important that they can recognize not just speech and music, which have been researched as specific tasks, but also general sounds in everyday environments. To stimulate research in this field we conducted a public research challenge: the IEEE Audio and Acoustic Signal Processing Technical Committee challenge on Detection and Classification of Acoustic Scenes and Events (DCASE). In this paper, we report on the state of the art in automatically classifying audio scenes, and automatically detecting and classifying audio events. We survey prior work as well as the state of the art represented by the submissions to the challenge from various research groups. We also provide detail on the organization of the challenge, so that our experience as challenge hosts may be useful to those organizing challenges in similar domains. We created new audio datasets and baseline systems for the challenge; these, as well as some submitted systems, are publicly available under open licenses, to serve as benchmarks for further research in general-purpose machine listening

Detection of overlapping acoustic events using a temporally-constrained probabilistic model

“…For constructing the pre-extracted dictionary P (f |q, c, s), the IEEE DCASE Event Detection training dataset was used [7,1]. The dataset contains isolated sounds recorded in an office environment at Queen Mary University of London, and covers 16 event classes (s ∈ {1, ..., 16}): alert, clearing throat, cough, door slam, drawer, keyboard click, keys, door knock, laughter, mouse click, page turn, pen drop, phone, printer, speech, and switch.…”

“…In addition, 3 monophonic datasets (1 real and 2 synthesized) of office sounds were also used, for comparative purposes. On the polyphonic datasets, firstly the test dataset for the IEEE DCASE Office Synthetic (OS) challenge was used [1]. The dataset contains 12 recordings of 2min duration each, with 3 different event density levels (low, mid, high) and 3 different SNR levels (-6dB,…”

“…For comparative purposes, 3 monophonic datasets of office sounds were used. Firstly, the Office Live (OL) dataset from the DCASE challenge was used [1], which contains 11 scripted recordings of event sequences recorded at Queen Mary University of London. The second and third monophonic datasets were generated using the acoustic scene synthesizer of [20], and each include 22 recordings of variable duration (1-3min), using as basis the annotations for the OL dataset.…”

“…For evaluation, the same set of metrics used for the IEEE DCASE event detection tasks was used [1]. Specifically, 3 different metrics are used: frame-based, event-based, and class-wise event-based.…”

“…The main goal of the aforementioned task is to label temporal regions within an audio recording, resulting in a symbolic description with start and end times, as well as labels for each instance of a specific event type [1]. Applications for acoustic event detection are numerous, including but not limited to security and surveillance, urban planning, smart homes, acoustic ecology, and organisation/navigation of sound archives [1,2,3,4].…”