IEEE AASP Challenge on Detection and Classification of Acoustic Scenes and Events

http://www.elec.qmul.ac.uk/digitalmusic/sceneseventschallenge/

Proposal Document: http://www.elec.qmul.ac.uk/digitalmusic/sceneseventschallenge/AASP_CASA.pdf

--

Development Dataset for Event Detection Task, subtask 1 - OL

Dataset developers: Dimitrios Giannoulis, Emmanouil Benetos, Dan Stowell, and Mark Plumbley (Queen Mary University of London)

--

The development dataset for the OL subtask consists of one file:

* events_OL_development.zip

The .zip file has the following folder structure:

* annotation1
* annotation2
* stereo
* bformat

--

The development dataset contains stereo and b-format recordings of scripted recordings containing non-overlapped acoustic events in an office environment. 
It contains events from 16 different classes, which are as follows:

* alarm (short alert (beep) sound)
* clearthroat (clearing throat)
* cough
* doorslam (door slam)
* drawer
* keyboard (keyboard clicks)
* keys (keys put on table)
* knock (door knock)
* laughter
* mouse (mouse click)
* pageturn (page turning)
* pendrop (pen, pencil, or marker touching table surfaces)
* phone
* printer
* speech
* switch

The recordings are made in a real working office environment, and so there are various (quiet) background sounds from nearby which have not been annotated and are to be treated as part of the background. 
An evident example is the sound of the door handle which has not been annotated as it is not part of the door slam sound which is one of the classes.

--

Folders 'annotation1' and 'annotation2' contain two different sets of annotations for the aforementioned recordings. The annotations for the first folder are named as 'scriptXX_bdm.txt' and for the second folder 'scriptXX_sid.txt' (please refer to the proposal document for additional information on the motivation behind two annotations). Each .txt file contains information about the onset, offset, and event class for each event present in the recording, and is structured as follows:

onset1	offset1	EventID1
onset2	offset2	EventID2
...

where 'onset' refers to the onset time in seconds, 'offset' refers to the offset time in seconds, and EventID is a string corresponding to the event class as shown in the list above. The three values are separated by tabs.

For the "printer" class, the onset-offset is annotated from when the printer actually starts printing to when it stops. Warming up and cooling down processes have not been accounted.

-- 

Folder 'stereo' contains the stero recordings for the scripted sequences. The recordings are named as scriptXX.wav, where XX is a 2-digit number.

--

Folder 'bformat' contains the original B-format recordings for the scripted sequences. Each recording includes 4 mono files, i.e. scriptXX-YY.wav, where YY={01,02,03,04}, corresponding to each SoundField channel. For information on recording specs, please refer to the challenge proposal document.

--

For any problems with this dataset, please contact: aasp-challenge-owner@eecs.qmul.ac.uk