TAU Urban Acoustic Scenes 2019 Openset, Development dataset

Audio Research Group / Tampere University

Authors

Recording and annotation

  • Henri Laakso
  • Ronal Bejarano Rodriguez
  • Toni Heittola

1. Dataset

TAU Urban Acoustic Scenes 2019 Openset development dataset is designed for researching open set classification problem in acoustic scene classification. In this problem, the test recording may be from a different environment than the 10 target classes. The dataset consists of 10-seconds audio segments from 10 acoustic scenes:

  • Airport - airport
  • Indoor shopping mall - shopping_mall
  • Metro station - metro_station
  • Pedestrian street - street_pedestrian
  • Public square - public_square
  • Street with medium level of traffic - street_traffic
  • Travelling by a tram - tram
  • Travelling by a bus - bus
  • Travelling by an underground metro - metro
  • Urban park - park

In addition, material for Unknown scene unknown is provided from 4 scenes:

  • Beach - beach, extracted from TUT Acoustic scenes 2017 dataset
  • Library - library, extracted from TUT Acoustic scenes 2017 dataset
  • Office - office, extracted from TUT Acoustic scenes 2017 dataset
  • Public event - public_event

Each target acoustic scene has 1440 segments (240 minutes of audio), unknown acoustic scene has 1450. The dataset contains in total 44 hours of audio.

The dataset was collected by Tampere University of Technology between 05/2018 - 11/2018. The data collection has received funding from the European Research Council under the ERC Grant Agreement 637422 EVERYSOUND.

ERC

Preparation of the dataset

The dataset was recorded in 12 large European cities: Amsterdam, Barcelona, Helsinki, Lisbon, London, Lyon, Madrid, Milan, Prague, Paris, Stockholm, and Vienna. For all acoustic scenes, audio was captured in multiple locations: different streets, different parks, different shopping malls. In each location, multiple 2-3 minute long audio recordings were captured in a few slightly different positions (2-4) within the selected location. Collected audio material was cut into segments of 10 seconds length.

The equipment used for recording consists of a binaural Soundman OKM II Klassik/studio A3 electret in-ear microphone and a Zoom F8 audio recorder using 48 kHz sampling rate and 24 bit resolution. During the recording, the microphones were worn by the recording person in the ears, and head movement was kept to minimum.

Post-processing of the recorded audio involves aspects related to privacy of recorded individuals, and possible errors in the recording process. The material was screened for content, and segments containing close microphone conversation were eliminated. Some interferences from mobile phones are audible, but are considered part of real-world recording process.

All provided audio data is single-channel, having a 44.1 KHz sampling rate, and 24 bit resolution.

A subset of the dataset has been previously published as TUT Urban Acoustic Scenes 2018 Development dataset. Audio segment filenames are retained for the segments coming from this dataset.

Dataset statistics

The development dataset contains audio material from 10 cities, whereas the evalution dataset (TAU Urban Acoustic Scenes 2019 Openset evaluation) contains data from all 12 cities.

The dataset is perfectly balanced at acoustic scene level, with very slight differences in the number of segments from each city.

Audio segments

| Scene class | Segments | Barcelona | Helsinki | Lisbon | London | Lyon | Milan | Paris | Prague | Stockholm | Vienna | DCASE2017 | | ------------------ | --------- | --------- | --------- | -------- | -------- | ---------| -------- | -------- | -------- | --------- | -------- | --------- | | Airport | 1440 | 128 | 149 | 144 | 145 | 144 | 144 | 156 | 144 | 158 | 128 | | | Bus | 1440 | 144 | 144 | 144 | 144 | 144 | 144 | 144 | 144 | 144 | 144 | | | Metro | 1440 | 141 | 144 | 144 | 146 | 144 | 144 | 144 | 144 | 145 | 144 | | | Metro station | 1440 | 144 | 144 | 144 | 144 | 144 | 144 | 144 | 144 | 144 | 144 | | | Park | 1440 | 144 | 144 | 144 | 144 | 144 | 144 | 144 | 144 | 144 | 144 | | | Public square | 1440 | 144 | 144 | 144 | 144 | 144 | 144 | 144 | 144 | 144 | 144 | | | Shopping mall | 1440 | 144 | 144 | 144 | 144 | 144 | 144 | 144 | 144 | 144 | 144 | | | Street, pedestrian | 1440 | 145 | 145 | 144 | 145 | 144 | 144 | 144 | 144 | 145 | 140 | | | Street, traffic | 1440 | 144 | 144 | 144 | 144 | 144 | 144 | 144 | 144 | 144 | 144 | | | Tram | 1440 | 143 | 145 | 144 | 144 | 144 | 144 | 144 | 144 | 144 | 144 | | | Unknown | 1450 | | | 134 |  | | | | 211 | | | 1105 | | Total | 15850 | 1421 | 1447 | 1574 | 1444 | 1440 | 1440 | 1452 | 1651 | 1456 | 1420 | 1105 |

Recording locations

Scene class Locations Barcelona Helsinki Lisbon London Lyon Milan Paris Prague Stockholm Vienna DCASE2017
Airport 40 4 3 4 3 4 4 4 6 5 3
Bus 71 4 4 11 7 7 7 11 10 6 4
Metro 67 3 5 11 4 9 8 9 10 4 4
Metro station 57 5 6 4 12 5 4 9 4 4 4
Park 41 4 4 4 4 4 4 4 4 5 4
Public_square 43 4 4 4 4 5 4 4 6 4 4
Shopping mall 36 4 4 4 2 3 3 4 4 4 4
Street, pedestrian 46 7 4 4 4 4 5 5 5 4 4
Street, traffic 43 4 4 4 5 4 6 4 4 4 4
Tram 70 4 4 6 9 7 11 9 11 5 4
Unknown 26 4 3 17
Total 514 43 42 56 54 52 56 63 65 45 39 17

File structure

dataset root
│   README.md               this file, markdown-format
│   README.html             this file, html-format
│   meta.csv                meta data, csv-format with a header row, [audio file (string)][tab][scene label (string)][tab][identifier (string)][tab][source_label (string)]
│   meta_unknown.csv        extra meta data for unknown class, csv-format with a header row, [audio file (string)][tab][scene label (string)][tab][identifier (string)][tab][source_label (string)][tab][original scene label (string)][tab][data source (string)]
│
└───audio                   15850 audio segments, 24-bit 44.1kHz mono
│   │   airport-barcelona-0-0-a.wav     file naming convention: [scene label]-[city]-[location id]-[segment id]-[device id].wav
│   │   airport-barcelona-0-1-a.wav
│   │   airport-barcelona-0-3-a.wav
│   │   ...
│   │   airport-barcelona-1-17-a.wav
│   │   airport-barcelona-1-18-a.wav
│   │   ...
│
└───evaluation_setup        cross-validation setup, 1 fold
    │   fold1_train.csv     training file list, csv-format with a header row, [audio file (string)][tab][scene label (string)]
    │   fold1_test.csv      testing file list, csv-format with a header row, [audio file (string)]
    │   fold1_evaluate.csv  evaluation file list, fold1_test.txt with added ground truth, csv-format with a header row, [audio file (string)][tab][scene label (string)]

2. Usage

The partitioning of the data was done based on the location of the original recordings. All segments recorded at the same location were included into a single subset - either development dataset or evaluation dataset. For each acoustic scene, 1440 segments were included in the development dataset provided here. Evaluation dataset is provided separately.

Training / test setup

A suggested training/test partitioning of the development set is provided in order to make results reported with this dataset uniform. The partitioning is done such that the segments recorded at the same location are included into the same subset - either training or testing. The partitioning is done aiming for a 70/30 ratio between the number of segments in training and test subsets while taking into account recording locations, and selecting the closest available option. Audio segments coming from nine cities are used for training and all ten cities are used for testing (Milan is used only for testing). Since the dataset includes balanced amount of material from ten cities, this partitioning will leave a small subset of data from Milan unused in the training / test setup. This material can be used when using full dataset to train the system and testing it with evaluation dataset.

Training examples for unknown scene class are all from TUT Acoustic scenes 2017 dataset, and testing examples are from same recording sessions as TAU Urban Acoustic Scenes 2019. Correspondence of unknown class examples with their original acoustic scenes and data source is provided in meta_unknown.csv.

The setup is provided with the dataset in the directory evaluation_setup.

Statistics

Scene class Train / Segments Train / Locations Test / Segments Test / Locations Unused / Segments Unused / Locations
Airport 911 25 421 12 108 3
Bus 928 46 415 20 97 5
Metro 902 41 433 20 105 6
Metro station 897 37 435 17 108 3
Park 946 27 386 11 108 3
Public square 945 28 387 12 108 3
Shopping mall 896 24 441 10 103 2
Street, pedestrian 924 29 429 14 87 3
Street, traffic 942 27 402 12 96 4
Tram 894 41 436 21 110 8
Unknown 1105 17 345 9
Total 10290 342 4530 158 1030 40

Training

evaluation setup\fold1_train.csv
training file list (in csv-format)

Format:

[audio file (string)][tab][scene label (string)]

Testing

evaluation setup\fold1_test.csv
testing file list (in csv-format)

Format: [audio file (string)]

Evaluating

evaluation setup\fold1_evaluate.csv
evaluation file list (in csv-format), same as fold1_test.csv but with additional reference information. These two files are provided separately to prevent contamination with ground truth when testing the system

Format: 

[audio file (string)][tab][scene label (string)]

Custom setups

If not using the provided training/test setup, pay attention to the segments recorded at the same location. Location identifier can be found from meta.csv or from audio file names:

[scene label]-[city]-[location id]-[segment id]-[device id].wav

Make sure that all files having same location id are placed on the same side of the evaluation. In this dataset, device id is always the same (a).

3. Changelog

v1.0 / 2019-03-11

  • Initial commit

4. License

License permits free academic usage. Any commercial use is strictly prohibited. For commercial use, contact dataset authors.

Copyright (c) 2019 Tampere University and its licensors
All rights reserved.
Permission is hereby granted, without written agreement and without license or royalty
fees, to use and copy the TAU Urban Acoustic Scenes 2019 Openset (“Work”) described in this document
and composed of audio and metadata. This grant is only for experimental and non-commercial
purposes, provided that the copyright notice in its entirety appear in all copies of this Work,
and the original source of this Work, (Audio Research Group at Tampere University of Technology),
is acknowledged in any publication that reports research using this Work.
Any commercial use of the Work or any part thereof is strictly prohibited.
Commercial use include, but is not limited to:
- selling or reproducing the Work
- selling or distributing the results or content achieved by use of the Work
- providing services by using the Work.

IN NO EVENT SHALL TAMPERE UNIVERSITY OR ITS LICENSORS BE LIABLE TO ANY PARTY
FOR DIRECT, INDIRECT, SPECIAL, INCIDENTAL, OR CONSEQUENTIAL DAMAGES ARISING OUT OF THE USE
OF THIS WORK AND ITS DOCUMENTATION, EVEN IF TAMPERE UNIVERSITY OR ITS
LICENSORS HAS BEEN ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

TAMPERE UNIVERSITY AND ALL ITS LICENSORS SPECIFICALLY DISCLAIMS ANY
WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND
FITNESS FOR A PARTICULAR PURPOSE. THE WORK PROVIDED HEREUNDER IS ON AN "AS IS" BASIS, AND
THE TAMPERE UNIVERSITY HAS NO OBLIGATION TO PROVIDE MAINTENANCE, SUPPORT,
UPDATES, ENHANCEMENTS, OR MODIFICATIONS.