Acoustic scene classification


Task description

The goal of acoustic scene classification is to classify a test recording into one of the predefined ten acoustic scene classes. This task is a continuation of the Acoustic Scene Classification task from previous DCASE Challenge editions, with some changes that bring new research problems into focus.

If you are interested in the task, you can join us on the dedicated slack channel

We provide two different setups of the acoustic classification problem:

A Complexity Task 1

Low-Complexity Acoustic Scene Classification with Multiple Devices
Subtask A

Classification of data from multiple devices (real and simulated) targeting generalization properties of systems across a number of different devices and focusing on low-complexity solutions.

B Modality Task 1

Audio-Visual Scene Classification
Subtask B

Classification of audio and video data, targeting learning of complementary information from different modalities, focusing on development of complex methods without restrictions on size or approach.

Subtask A

A Complexity Task 1

Low-Complexity Acoustic Scene Classification with Multiple Devices

This subtask is concerned with the basic problem of acoustic scene classification, in which it is required to classify a test audio recording into one of ten known acoustic scene classes. This task targets generalization across a number of different devices, and will use audio data recorded and simulated with a variety of devices. The task also targets low complexity solutions for the classification problem in terms of model size.

Figure 1: Overview of acoustic scene classification system.


Audio dataset

The development dataset for this task is TAU Urban Acoustic Scenes 2020 Mobile, development dataset. The dataset contains recordings from 12 European cities in 10 different acoustic scenes using 4 different devices. Additionally, synthetic data for 11 mobile devices was created based on the original recordings. Of the 12 cities, two are present only in the evaluation set.

Recordings were made using four devices that captured audio simultaneously. The main recording device consists in a Soundman OKM II Klassik/studio A3, electret binaural microphone and a Zoom F8 audio recorder using 48kHz sampling rate and 24-bit resolution, referred to as device A. The other devices are commonly available customer devices: device B is a Samsung Galaxy S7, device C is iPhone SE, and device D is a GoPro Hero5 Session.

Acoustic scenes (10):

  • 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

Audio data was recorded in Amsterdam, Barcelona, Helsinki, Lisbon, London, Lyon, Madrid, Milan, Prague, Paris, Stockholm and Vienna.

The dataset was collected by Tampere University of Technology between 05/2018 - 11/2018. The data collection received funding from the European Research Council, grant agreement 637422 EVERYSOUND.

ERC

For complete details on the data recording and processing see

Publication

Annamaria Mesaros, Toni Heittola, and Tuomas Virtanen. A multi-device dataset for urban acoustic scene classification. In Proceedings of the Detection and Classification of Acoustic Scenes and Events 2018 Workshop (DCASE2018), 9–13. November 2018. URL: https://arxiv.org/abs/1807.09840.

PDF

A multi-device dataset for urban acoustic scene classification

Abstract

This paper introduces the acoustic scene classification task of DCASE 2018 Challenge and the TUT Urban Acoustic Scenes 2018 dataset provided for the task, and evaluates the performance of a baseline system in the task. As in previous years of the challenge, the task is defined for classification of short audio samples into one of predefined acoustic scene classes, using a supervised, closed-set classification setup. The newly recorded TUT Urban Acoustic Scenes 2018 dataset consists of ten different acoustic scenes and was recorded in six large European cities, therefore it has a higher acoustic variability than the previous datasets used for this task, and in addition to high-quality binaural recordings, it also includes data recorded with mobile devices. We also present the baseline system consisting of a convolutional neural network and its performance in the subtasks using the recommended cross-validation setup.

Keywords

Acoustic scene classification, DCASE challenge, public datasets, multi-device data

PDF

Additionally, 11 mobile devices S1-S11 are simulated using the audio recorded with device A, impulse responses recorded with real devices, and additional dynamic range compression, in order to simulate realistic recordings. A recording from device A is processed through convolution with the selected Si impulse response, then processed with a selected set of parameters for dynamic range compression (device-specific). The impulse responses are proprietary data and will not be published.

The development dataset comprises 40 hours of data from device A, and smaller amounts from the other devices. Audio is provided in a single-channel 44.1kHz 24-bit format.

Task setup

Development dataset

The development set contains data from 10 cities and 9 devices: 3 real devices (A, B, C) and 6 simulated devices (S1-S6). Data from devices B, C, and S1-S6 consists of randomly selected segments from the simultaneous recordings, therefore all overlap with the data from device A, but not necessarily with each other. The total amount of audio in the development set is 64 hours.

The dataset is provided with a training/test split in which 70% of the data for each device is included for training, 30% for testing. Some devices appear only in the test subset. In order to create a perfectly balanced test set, a number of segments from various devices are not included in this split. Complete details on the development set and training/test split are provided in the following table.

Devices Dataset Cross-validation setup
Name Type Total
duration
Total
segments
Train
segments
Test
segments
Notes
A Real 40h 14400 10215 330 3855 Segments not used in train/test split
B C Real 3h each 1080 749 + 748 2 * 329
S1 S2 S3 Simulated 3h each 1080 3 * 750 2 * 330
S4 S5 S6 Simulated 3h each 1080 - 3 * 330 2 * 750 segments not used in train/test split
Total 64h 23040 13962 2970

Participants are required to report the performance of their system using this train/test setup in order to allow a comparison of systems on the development set. Participants are allowed to create their own cross-validation folds or separate validation set. In this case please pay attention to the segments recorded at the same location. Location identifier can be found from metadata file provided in the dataset or from audio file names:

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

Make sure that all the files having the same location id are placed on the same side of the evaluation.

Evaluation dataset

The evaluation dataset contains data from 12 cities, 10 acoustic scenes, 11 devices. There are five new devices (not available in the development set): a real device D and simulated devices S7-S11. Evaluation data contains 33 hours of audio. The evaluation data contains audio recorded at different locations than the development data.

Device and city information is not provided in the evaluation set. The systems are expected to be robust to different devices.

System complexity requirements

A model complexity limit of 128 KB is set for the non-zero parameters. This translates into 32768 parameters when using float32 (32-bit float) which is often the default data type (32768 parameter values * 32 bits per parameter / 8 bits per byte= 131072 bytes = 128 KB (kibibyte)).

By limiting the size of the model on disk, we allow participants some flexibility in design, for example, some implementations would prefer to minimize the number of non-zero parameters of the network in order to comply with this size limit, while other implementations may target representation of the model parameters with a low number of bits. There is no requirement nor recommendation on which method to minimize the model size is sought after.

The computational complexity of the feature extraction stage is not included in this limit because there is no established method for estimating and comparing complexity of different low-level feature extraction implementations. We therefore exclude it in order to keep the complexity estimation straightforward across approaches. Some implementations may use a feature extraction layer as the first layer in the neural network - in this case the limit is applied only to the following layers, in order to exclude the feature calculation as if it were a separate processing block. However, in case of using learned features (so-called embeddings, like VGGish, OpenL3 or EdgeL3), the network used to generate them counts in the calculated model size.

Full information about the model size should be provided in the technical report.

Model size calculation

We offer a script for calculating the model size for Keras based models along with the baseline system. If you have any doubts about how to calculate the model size, please contact toni.heittola@tuni.fi or use the DCASE community forum or Slack channel for visibility.

Calculation examples

Total model size: 17.89 MB (Audio embeddings) + 1.254 MB (Acoustic model) = 19.14 MB

Audio embeddings (OpenL3)
Layer Parameters Non-zero parameters Data type Size (non-zero) Note
input_1 0 0 float32 0 KB
melspectrogram_1 4 460 800 4 196 335 float32 16.01 MB Skipped
batch_normalization_1 4 4 float32 16 bytes
conv2d_1 640 640 float32 2.5 KB
batch_normalization_2 256 256 float32 1 KB
activation_1 0 0 float32 0 KB
conv2d_2 36 928 36 928 float32 144.2 KB
batch_normalization_3 256 256 float32 1 KB
activation_2 0 0 float32 0 KB
max_pooling2d_1 0 0 float32 0 KB
conv2d_3 73 856 73 856 float32 288.5 KB
batch_normalization_4 512 512 float32 2 KB
activation_3 0 0 float32 0 KB
conv2d_4 147 584 147 584 float32 576.5 KB
batch_normalization_5 512 512 float32 2 KB
activation_4 0 0 float32 0 KB
max_pooling2d_2 0 0 float32 0 KB
conv2d_5 295 168 295 168 float32 1.126 MB
batch_normalization_6 1024 1024 float32 4 KB
activation_5 0 0 float32 0 KB
conv2d_6 590 080 590 080 float32 2.251 MB
batch_normalization_7 1024 1024 float32 4 KB
activation_6 0 0 float32 0 KB
max_pooling2d_3 0 0 float32 0 KB
conv2d_7 1 180 160 1 180 160 float32 4.502 MB
batch_normalization_8 2048 2048 float32 8 KB
activation_7 0 0 float32 0 KB
audio_embedding_layer 2 359 808 2 359 808 float32 9.002 MB
max_pooling2d_4 0 0 float32 0 KB
flatten_1 0 0 float32 0 KB
Total 4 689 860 4 689 860 17.89 MB
(4689860 * 32bit / 8bits per byte / 1024 / 1024)
Acoustic model
Layer Parameters Non-zero parameters Size (non-zero) Note
dense_1 262 656 262 557 1.002 MB
dense_2 65 664 65 664 256.5 KB
dense_3 387 387 1.512 KB
Total 32 8707 32 8608 1.254 MB

Total model size: 0 KB (Audio embeddings) + 451.5 KB (Acoustic model) = 451.5 KB

Acoustic model
Layer Parameters Non-zero parameters Data type Size (non-zero) Note
conv2d_1 1600 1600 float32 6.25 KB
batch_normalization_1 128 128 float32 512 bytes
activation_1 0 0 float32 0 KB
max_pooling2d_1 0 0 float32 0 KB
dropout_1 0 0 float32 0 KB
conv2d_2 100 416 100 416 float32 392.2 KB
batch_normalization_2 256 256 float32 1 KB
activation_2 0 0 float32 0 KB
max_pooling2d_2 0 0 float32 0 KB
dropout_2 0 0 float32 0 KB
flatten_1 0 0 float32 0 KB
dense_1 12 900 12 900 float32 50.39 KB
dropout_3 0 0 float32 0 KB
dense_2 303 303 float32 1.184 KB
Total 115 603 115 603 451.5 KB
(115603 * 32bit / 8bits per byte / 1024)

Total model size: 0 KB (Audio embeddings) + 90.3 KB (Acoustic model) = 90.3 KB

Acoustic model
Layer Parameters Non-zero parameters Data type Size (non-zero) Note
conv2d_1 800 800 float16 1.56 KB
batch_normalization_1 64 64 float16 128 bytes
activation_1 0 0 0 KB
conv2d_2 12 560 12 560 float16 24.53 KB
batch_normalization_2 64 64 float16 128 bytes
activation_2 0 0 0 KB
max_pooling2d_1 0 0 0 KB
dropout_1 0 0 0 KB
conv2d_3 25 120 25 120 float16 49.06 KB
batch_normalization_3 128 128 float16 256 bytes
activation_3 0 0 0 KB
max_pooling2d_2 0 0 0 KB
dropout_2 0 0 0 KB
flatten_1 0 0 0 KB
dense_1 6 500 6 500 float16 12.69 KB
dropout_3 0 0 0 KB
dense_2 1010 1010 float16 1.97 KB
Total 46 246 46 246 90.3 KB
(46246 * 16bit / 8bits per byte / 1024)

Download


Subtask B

B Modality Task 1

Audio-Visual Scene Classification
Subtask B

This subtask is concerned with classification using audio and video modalities. Since audio-visual machine learning has gained popularity in the last years, we aim to provide a multidisciplinary task that may attract researchers from the machine vision community.

We impose no restrictions on the modality or combinations of modalities used in the system. We encourage participants to also submit single-modality systems (audio-only or video-only methods for scene classification).

Figure 2: Overview of audio-visual scene classification system.


Audio-Visual dataset

The dataset for this task is TAU Audio-Visual Urban Scenes 2021. The dataset contains synchronized audio and video recordings from 12 European cities in 10 different scenes.

The audio part is a subset of TAU Urban Acoustic Scenes 2020. For complete details on the data recording and processing see:

Publication

Shanshan Wang, Annamaria Mesaros, Toni Heittola, and Tuomas Virtanen. A curated dataset of urban scenes for audio-visual scene analysis. In 2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2021. accepted. URL: https://arxiv.org/abs/2011.00030.

PDF

A Curated Dataset of Urban Scenes for Audio-Visual Scene Analysis

Abstract

This paper introduces a curated dataset of urban scenes for audio-visual scene analysis which consists of carefully selected and recorded material. The data was recorded in multiple European cities, using the same equipment, in multiple locations for each scene, and is openly available. We also present a case study for audio-visual scene recognition and show that joint modeling of audio and visual modalities brings significant performance gain compared to state of the art uni-modal systems. Our approach obtained an 84.4% accuracy compared to 76.8% for the audio-only and 70.0% for the video-only equivalent systems.

Keywords

Audio-visual data, Scene analysis, Acous-tic scene, Pattern recognition, Transfer learning

PDF

The provided audio is recorded using a Soundman OKM II Klassik/studio A3, electret binaural microphone and a Zoom F8 audio recorder using 48kHz sampling rate and 24-bit resolution. The provided video is recorded using a GoPro Hero5 Session. Faces and licence plates in the video were blurred during the data postprocessing stage.

Data was recorded in the following scenes (10):

  • 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

Data was recorded in Amsterdam, Barcelona, Helsinki, Lisbon, London, Lyon, Madrid, Milan, Prague, Paris, Stockholm and Vienna.

The dataset was collected by Tampere University of Technology between 05/2018 - 11/2018. The data collection received funding from the European Research Council, grant agreement 637422 EVERYSOUND.

ERC

Task setup

Development dataset

The development set contains audio and video data from 10 cities. The total amount of audio in the development set is 34 hours. The dataset is provided with a training/test split.

Provided files have a length of 10 seconds. A classification decision is required for 1 second segments. Participants are allowed to implement this in any way they want, by splitting the 10-second files into individual 1-second files, or just providing 10 labels independently within the longer file. In the baseline system, the latter method is used. Participants are required to report the performance of their system using this train/test setup in order to allow comparison of systems on the development set.

Participants are allowed to create their own cross-validation folds or separate validation set. In this case please pay attention to the segments recorded at the same location. Location identifier can be found from metadata file provided in the dataset or from audio file names:

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

Make sure that all files having the same location id are placed on the same side of the evaluation.

Evaluation dataset

The evaluation set contains data from 12 cities (2 cities unseen in the development set). Evaluation data contains 20 hours of material (audio and video). Provided files have a length of 1 second. Classification decision is required at file level.

Download


External data resources

Use of external data and transfer learning is allowed in all subtasks under the following conditions:

  • The used external resource is clearly referenced and freely accessible to any other research group in the world. External data refers to public datasets or trained models. The data must be public and freely available before 1st of April 2021.

  • The list of external data sources used in training must be clearly indicated in the technical report.

  • Participants inform the organizers in advance about such data sources, so that all competitors know about them and have an equal opportunity to use them. Please send an email to the task coordinators; we will update the list of external datasets on the webpage accordingly. Once the evaluation set is published, the list of allowed external data resources is locked (no further external sources allowed).

  • It is not allowed to use TUT Urban Acoustic Scenes 2018, TAU Urban Acoustic Scenes 2019, TAU Urban Acoustic Scenes 2019 Mobile, TAU Urban Acoustic Scenes 2020 Mobile, or TAU Urban Acoustic Scenes 2020 3Class. These datasets are partially included in the current setup, and additional usage will lead to overfitting.

List of external data resources allowed:

Dataset name Type Added Link
LITIS Rouen audio scene dataset audio 04.03.2019 https://sites.google.com/site/alainrakotomamonjy/home/audio-scene
DCASE2013 Challenge - Public Dataset for Scene Classification Task audio 04.03.2019 https://archive.org/details/dcase2013_scene_classification
DCASE2013 Challenge - Private Dataset for Scene Classification Task audio 04.03.2019 https://archive.org/details/dcase2013_scene_classification_testset
AudioSet audio, video 04.03.2019 https://research.google.com/audioset/
OpenL3 model 12.02.2020 https://openl3.readthedocs.io/
EdgeL3 model 12.02.2020 https://edgel3.readthedocs.io/
VGGish model 12.02.2020 https://github.com/tensorflow/models/tree/master/research/audioset/vggish
SoundNet model 03.06.2020 http://soundnet.csail.mit.edu/
CIFAR-100 image 1.3.2021 https://www.cs.toronto.edu/~kriz/cifar.html
CIFAR-10 image 31.3.2021 https://www.cs.toronto.edu/~kriz/cifar.html
ImageNet image 1.3.2021 http://www.image-net.org/
Resnet50 model 1.3.2021 https://pytorch.org/hub/pytorch_vision_resnet/
EfficientNet model 1.3.2021 https://github.com/lukemelas/EfficientNet-PyTorch
Indoor image 18.3.2021 http://web.mit.edu/torralba/www/indoor.html
Places365 image 18.3.2021 http://places2.csail.mit.edu/download.html
Urban-SED audio 31.3.2021 http://urbansed.weebly.com/
Pytorch CIFAR Models model 31.3.2021 https://github.com/chenyaofo/pytorch-cifar-models
Places365-CNNs model 31.3.2021 https://github.com/CSAILVision/places365
Pytorch pretrained Models on ImageNet model 31.3.2021 https://pytorch.org/vision/stable/models.html
PANNs: Large-Scale Pretrained Audio Neural Networks for Audio Pattern Recognition model 31.3.2021 https://zenodo.org/record/3987831
Problem Agnostic Speech Encoder (PASE) Model model 31.3.2021 https://github.com/santi-pdp/pase


Task rules

There are general rules valid for all tasks; these, along with information on technical report and submission requirements can be found here.

Task specific rules:

  • Use of external data is allowed. See conditions for external data usage here.
  • In subtask A, the model size limit applies. See conditions for the model size here.
  • Manipulation of provided training and development data is allowed (e.g. by mixing data sampled from a pdf or using techniques such as pitch shifting or time stretching).
  • Participants are not allowed to make subjective judgments of the evaluation data, nor to annotate it. The evaluation dataset cannot be used to train the submitted system; the use of statistics about the evaluation data in the decision making is also forbidden.
  • Classification decision must be done independently for each test sample.

Submission

Participants can choose to participate in only one subtask or both. For Subtask B, any combination of modalities can be submitted (audio-only, video-only, audio-video)

Official challenge submission consists of:

  • System output file (*.csv)

  • Metadata file (*.yaml)

  • Technical report explaining in sufficient detail the method (*.pdf)

System output should be presented as a single text-file (in CSV format, with a header row) containing a classification result for each audio file in the evaluation set. In addition, the results file should contain probabilities for each scene class. Result items can be in any order. Multiple system outputs can be submitted (maximum 4 per participant per subtask).

For each system, meta information should be provided in a separate file, containing the task-specific information. This meta information enables fast processing of the submissions and analysis of submitted systems. Participants are advised to fill the meta information carefully while making sure all information is correctly provided.

All files should be packaged into a zip file for submission. Please make a clear connection between the system name in the submitted yaml, submitted system output, and the technical report! Instead of system name you can use a submission label too.

System output file

Both subtask will follow the same system output file format.

Row format:

[filename (string)][tab][scene label (string)][tab][bus probability (float)][tab][metro probability (float)][tab][metro_station probability (float)][tab][park probability (float)][tab][public_square probability (float)][tab][shopping_mall probability (float)][tab][street_pedestrian probability (float)][tab][street_traffic probability (float)][tab][tram probability (float)]

Example output:

filename	scene_label	airport	bus	metro	metro_station	park	public_square	shopping_mall	street_pedestrian	street_traffic	tram
0.wav	bus	0.25	0.99	0.12	0.32	0.41	0.42	0.23	0.34	0.12	0.45
1.wav	tram	0.25	0.19	0.12	0.32	0.41	0.42	0.23	0.34	0.12	0.85

Metadata file

Subtask A

Example meta information file for Subtask A baseline system task1/Martin_TAU_task1a_1/Martin_TAU_task1a_1.meta.yaml:

# Submission information
submission:
  # Submission label
  # Label is used to index submissions.
  # Generate your label following way to avoid
  # overlapping codes among submissions:
  # [Last name of corresponding author]_[Abbreviation of institute of the corresponding author]_task[task number]_[index number of your submission (1-4)]
  label: Martin_TAU_task1a_1

  # Submission name
  # This name will be used in the results tables when space permits
  name: DCASE2021 baseline system

  # Submission name abbreviated
  # This abbreviated name will be used in the results table when space is tight.
  # Use maximum 10 characters.
  abbreviation: Baseline

  # Authors of the submitted system. Mark authors in
  # the order you want them to appear in submission lists.
  # One of the authors has to be marked as corresponding author,
  # this will be listed next to the submission in the results tables.
  authors:
    # First author
    - lastname: Martín Morató
      firstname: Irene
      email: irene.martinmorato@tuni.fi           # Contact email address
      corresponding: true                         # Mark true for one of the authors

      # Affiliation information for the author
      affiliation:
        abbreviation: TAU
        institute: Tampere University
        department: Computing Sciences            # Optional
        location: Tampere, Finland

    # Second author
    - lastname: Heittola
      firstname: Toni
      email: toni.heittola@tuni.fi                # Contact email address

      # Affiliation information for the author
      affiliation:
        abbreviation: TAU
        institute: Tampere University
        department: Computing Sciences            # Optional
        location: Tampere, Finland

    # Third author
    - lastname: Mesaros
      firstname: Annamaria
      email: annamaria.mesaros@tuni.fi

      # Affiliation information for the author
      affiliation:
        abbreviation: TAU
        institute: Tampere University
        department: Computing Sciences
        location: Tampere, Finland

    # Fourth author
    - lastname: Virtanen
      firstname: Tuomas
      email: tuomas.virtanen@tuni.fi

      # Affiliation information for the author
      affiliation:
        abbreviation: TAU
        institute: Tampere University
        department: Computing Sciences
        location: Tampere, Finland

# System information
system:
  # System description, meta data provided here will be used to do
  # meta analysis of the submitted system.
  # Use general level tags, when possible use the tags provided in comments.
  # If information field is not applicable to the system, use "!!null".
  description:

    # Audio input / sampling rate
    # e.g. 16kHz, 22.05kHz, 44.1kHz, 48.0kHz
    input_sampling_rate: 44.1kHz

    # Acoustic representation
    # one or multiple labels, e.g. MFCC, log-mel energies, spectrogram, CQT, raw waveform, ...
    acoustic_features: log-mel energies

    # Embeddings
    # e.g. VGGish, OpenL3, ...
    embeddings: !!null

    # Data augmentation methods
    # e.g. mixup, time stretching, block mixing, pitch shifting, ...
    data_augmentation: !!null

    # Machine learning
    # In case using ensemble methods, please specify all methods used (comma separated list).
    # one or multiple, e.g. GMM, HMM, SVM, MLP, CNN, RNN, CRNN, ResNet, ensemble, ...
    machine_learning_method: CNN

    # Ensemble method subsystem count
    # In case ensemble method is not used, mark !!null.
    # e.g. 2, 3, 4, 5, ...
    ensemble_method_subsystem_count: !!null

    # Decision making methods
    # e.g. average, majority vote, maximum likelihood, ...
    decision_making: !!null

    # External data usage method
    # e.g. directly, embeddings, pre-trained model, ...
    external_data_usage: embeddings

    # Method for handling the complexity restrictions
    # e.g. weight quantization, sparsity, ...
    complexity_management: weight quantization

  # System complexity, meta data provided here will be used to evaluate
  # submitted systems from the computational load perspective.
  complexity:
    # Total amount of parameters used in the acoustic model.
    # For neural networks, this information is usually given before training process
    # in the network summary.
    # For other than neural networks, if parameter count information is not directly
    # available, try estimating the count as accurately as possible.
    # In case of ensemble approaches, add up parameters for all subsystems.
    # In case embeddings are used, add up parameter count of the embedding
    # extraction networks and classification network
    # Use numerical value.
    total_parameters: 46246

    # Total amount of non-zero parameters in the acoustic model.
    # Calculated with same principles as "total_parameters".
    # Use numerical value.
    total_parameters_non_zero: 46246

    # Model size calculated as instructed in task description page.
    # Use numerical value, unit is KB
    model_size: 90.3 # KB

  # List of external datasets used in the submission.
  # Development dataset is used here only as example, list only external datasets
  external_datasets:
    # Dataset name
    - name: TAU Urban Acoustic Scenes 2020 Mobile, Development dataset

      # Dataset access url
      url: https://doi.org/10.5281/zenodo.3819968

      # Total audio length in minutes
      total_audio_length: 3840            # minutes

  # URL to the source code of the system [optional]
  source_code: https://github.com/marmoi/dcase2021_task1a_baseline

# System results
results:
  development_dataset:
    # System results for development dataset with provided the cross-validation setup.
    # Full results are not mandatory, however, they are highly recommended
    # as they are needed for through analysis of the challenge submissions.
    # If you are unable to provide all results, also incomplete
    # results can be reported.

    # Overall metrics
    overall:
      logloss: 1.461
      accuracy: 46.9    # mean of class-wise accuracies

    # Class-wise metrics
    class_wise:
      airport:
        logloss: 1.497
        accuracy: 31.1
      bus:
        logloss: 1.475
        accuracy: 40.1
      metro:
        logloss: 1.457
        accuracy: 48.1
      metro_station:
        logloss: 2.060
        accuracy: 29.6
      park:
        logloss: 1.217
        accuracy: 63.6
      public_square:
        logloss: 1.738
        accuracy: 36.0
      shopping_mall:
        logloss: 1.136
        accuracy: 61.3
      street_pedestrian:
        logloss: 1.522
        accuracy: 47.1
      street_traffic:
        logloss: 1.145
        accuracy: 68.0
      tram:
        logloss: 1.360
        accuracy: 44.3

    # Device-wise
    device_wise:
      a:
        logloss: !!null
        accuracy: 63.9
      b:
        logloss: !!null
        accuracy: 52.2
      c:
        logloss: !!null
        accuracy: 56.3
      s1:
        logloss: !!null
        accuracy: 44.2
      s2:
        logloss: !!null
        accuracy: 43.9
      s3:
        logloss: !!null
        accuracy: 44.5
      s4:
        logloss: !!null
        accuracy: 38.5
      s5:
        logloss: !!null
        accuracy: 40.6
      s6:
        logloss: !!null
        accuracy: 38.2

Subtask B

Example meta information file for subtask B baseline system task1/Wang_TAU_task1b_1/Wang_TAU_task1b_1.meta.yaml:

# Submission information
submission:
  # Submission label
  # Label is used to index submissions.
  # Generate your label following way to avoid
  # overlapping codes among submissions:
  # [Last name of corresponding author]_[Abbreviation of institute of the corresponding author]_task[task number]_[index number of your submission (1-4)]
  label: Wang_TAU_task1b_1

  # Submission name
  # This name will be used in the results tables when space permits
  name: DCASE2021 baseline system

  # Submission name abbreviated
  # This abbreviated name will be used in the results table when space is tight.
  # Use maximum 10 characters.
  abbreviation: Baseline

  # Authors of the submitted system. Mark authors in
  # the order you want them to appear in submission lists.
  # One of the authors has to be marked as corresponding author,
  # this will be listed next to the submission in the results tables.
  authors:
    # First author
    - lastname: Wang
      firstname: Shanshan
      email: shanshan.wang@tuni.fi                # Contact email address
      corresponding: true                         # Mark true for one of the authors

      # Affiliation information for the author
      affiliation:
        abbreviation: TAU
        institute: Tampere University
        department: Computing Sciences            # Optional
        location: Tampere, Finland

    # Second author
    - lastname: Heittola
      firstname: Toni
      email: toni.heittola@tuni.fi                # Contact email address

      # Affiliation information for the author
      affiliation:
        abbreviation: TAU
        institute: Tampere University
        department: Computing Sciences            # Optional
        location: Tampere, Finland

    # Third author
    - lastname: Mesaros
      firstname: Annamaria
      email: annamaria.mesaros@tuni.fi

      # Affiliation information for the author
      affiliation:
        abbreviation: TAU
        institute: Tampere University
        department: Computing Sciences
        location: Tampere, Finland

    # Fourth author
    - lastname: Virtanen
      firstname: Tuomas
      email: tuomas.virtanen@tuni.fi

      # Affiliation information for the author
      affiliation:
        abbreviation: TAU
        institute: Tampere University
        department: Computing Sciences
        location: Tampere, Finland

# System information
system:
  # System description, meta data provided here will be used to do
  # meta analysis of the submitted system.
  # Use general level tags, when possible use the tags provided in comments.
  # If information field is not applicable to the system, use "!!null".
  description:
    # Audio input / channels
    # one or multiple: e.g. mono, binaural, left, right, mixed, ...
    input_channels: mono

    # Audio input / sampling rate
    # e.g. 16kHz, 22.05kHz, 44.1kHz, 48.0kHz
    input_sampling_rate: 48.0kHz

    # Acoustic representation
    # one or multiple labels, e.g. MFCC, log-mel energies, spectrogram, CQT, raw waveform, ...
    acoustic_features: log-mel energies

    # Embeddings
    # e.g. VGGish, OpenL3, ...
    audio_embeddings: OpenL3
    visual_embeddings: OpenL3

    # Data augmentation methods
    # e.g. mixup, time stretching, block mixing, pitch shifting, ...
    data_augmentation: !!null

    # Machine learning
    # In case using ensemble methods, please specify all methods used (comma separated list).
    # one or multiple, e.g. GMM, HMM, SVM, MLP, CNN, RNN, CRNN, ResNet, ensemble, ...
    machine_learning_method: CNN

    # Ensemble method subsystem count
    # In case ensemble method is not used, mark !!null.
    # e.g. 2, 3, 4, 5, ...
    ensemble_method_subsystem_count: !!null

    # How information from modalities are combined
    # e.g. audio only, video only, early fusion, late fusion
    modality_combination: early fusion

    # Decision making methods
    # e.g. average, majority vote, maximum likelihood, ...
    decision_making: !!null

    # External data usage method
    # e.g. directly, embeddings, pre-trained model, ...
    external_data_usage: embeddings

  # System complexity, meta data provided here will be used to evaluate
  # submitted systems from the computational load perspective.
  complexity:
    # Total amount of parameters used in the model.
    # For neural networks, this information is usually given before training process
    # in the network summary.
    # For other than neural networks, if parameter count information is not directly
    # available, try estimating the count as accurately as possible.
    # In case of ensemble approaches, add up parameters for all subsystems.
    # In case embeddings are used, add up parameter count of the embedding
    # extraction networks and classification network
    # Use numerical value.
    total_parameters: 115219

    # Amount of parameters used in the acoustic model. Indicated the same way than total_parameters.
    # Use numerical value.
    total_parameters_audio: 0

    # Amount of parameters used in the visual model. Indicated the same way than total_parameters
    # Use numerical value.
    total_parameters_visual: 0

  # List of external datasets used in the submission.
  # Development dataset is used here only as example, list only external datasets
  external_datasets:
    # Dataset name
    - name: TAU Urban Audio-Visual Scenes 2021, Development dataset

      # Dataset access url
      url:

      # Total audio length in minutes
      total_audio_length: 2040            # minutes

  # URL to the source code of the system [optional]
  source_code: https://github.com/shanwangshan/TAU-urban-audio-visual-scenes

# System results
results:
  development_dataset:
    # System results for development dataset with provided the cross-validation setup.
    # Full results are not mandatory, however, they are highly recommended
    # as they are needed for through analysis of the challenge submissions.
    # If you are unable to provide all results, also incomplete
    # results can be reported.

    # Overall metrics
    overall:
      logloss: 0.658
      accuracy: 77.0    # mean of class-wise accuracies


    # Class-wise metrics
    class_wise:
      airport:
        logloss: !!null
        accuracy: 66.8
      bus:
        logloss: !!null
        accuracy: 85.9
      metro:
        logloss: !!null
        accuracy: 80.4
      metro_station:
        logloss: !!null
        accuracy: 80.8
      park:
        logloss: !!null
        accuracy: 77.2
      public_square:
        logloss: !!null
        accuracy: 71.1
      shopping_mall:
        logloss: !!null
        accuracy: 72.6
      street_pedestrian:
        logloss: !!null
        accuracy: 72.7
      street_traffic:
        logloss: !!null
        accuracy: 89.6
      tram:
        logloss: !!null
        accuracy: 73.1

Package validator

An automatic validation tool to help challenge participants to prepare a correctly formatted submission package will be released along with the evaluation datasets. Properly validated submission package will speed up the submission processing in the challenge evaluation stage. Please use this to make sure your submission package follows the given formatting.

Evaluation

Systems will be ranked by macro-average multiclass cross-entropy (Log loss) (average of the class-wise log loss). The metric is independent of the operating point (see python implementation here).

As a secondary metric, we will calculate macro-average accuracy (average of the class-wise accuracies).

Baseline systems

The baseline system provides a state-of-the-art approach for the classification in each subtask.

Subtask A

The baseline system implements a convolutional neural network (CNN) based approach using log mel-band energies extracted for each 10-second signal. The network consists of three CNN layers and one fully connected layer to assign scene labels to the audio signals. The system is based on the DCASE 2020 Subtask B baseline system.

After training, the model parameters are quantized to float16. The resulting model size is 90.3 KB.

Repository


Parameters

  • Audio features:
    • Log mel-band energies (40 bands), analysis frame 40 ms (50% hop size)
  • Neural network:
    • Input shape: 40 * 500 (10 seconds)
    • Architecture:
      • CNN layer #1: 2D Convolutional layer (filters: 16, kernel size: 7) + Batch normalization + ReLu activation
      • CNN layer #2: 2D Convolutional layer (filters: 16, kernel size: 7) + Batch normalization + ReLu activation, 2D max pooling (pool size: (5, 5)) + Dropout (rate: 30%)
      • CNN layer #3: 2D Convolutional layer (filters: 32, kernel size: 7) + Batch normalization + ReLu activation, 2D max pooling (pool size: (4, 100)) + Dropout (rate: 30%)
      • Flatten
      • Dense layer #1: Dense layer (units: 100, activation: ReLu ), Dropout (rate: 30%)
      • Output layer (activation: softmax)
    • Learning: 200 epochs (batch size 16), data shuffling between epochs
    • Optimizer: Adam (learning rate 0.001)

Model selection:

  • Approximately 30% of the original training data is assigned to validation set, split done such that training and validation sets do not have segments from the same location and both sets have data from each city
  • Model performance after each epoch is evaluated on the validation set, and best performing model is selected

Results for the development dataset

System Log loss Accuracy Description
DCASE2021 Task 1 Baseline, Subtask A 1.551
(± 0.036)
46.4 %
(± 0.9)
Log mel-band energies as features,
three layers of 2D CNN and one fully connected layer as classifier. After training, the model parameters are quantized to float16.
DCASE2020 Task 1 Baseline, Subtask A 1.365
(± 0.032)
54.1 %
(± 1.4)
OpenL3 as audio embeddings,
two fully connected layers as classifiers
DCASE2019 Task 1 Baseline 1.578
(± 0.029)
46.5 %
(± 1.2)
Log mel-band energies as features,
two layers of 2D CNN and one fully connected layer as classifier,

See more information

Results for DCASE2021 baseline are calculated using TensorFlow in GPU mode (using Nvidia Tesla V100 GPU card). Because results produced with GPU card are generally non-deterministic, the system was trained and tested 10 times; mean and standard deviation of the performance from these 10 independent trials are shown in the results tables. Detailed results for the DCASE2021 baseline:

Scene label Log loss Device-wise log-losses Accuracy
A B C S1 S2 S3 S4 S5 S6
Airport 1.758 1.031 1.687 1.743 1.858 1.498 1.798 1.975 1.754 2.474 24.7 %
Bus 1.170 0.712 1.538 0.838 1.003 1.374 0.807 1.566 1.153 1.539 58.6 %
Metro 1.719 0.895 1.726 1.204 2.750 2.06 1.591 1.439 1.833 1.974 37.4 %
Metro station 1.859 1.502 1.844 2.275 2.342 2.131 1.704 1.872 1.671 1.390 38.7 %
Park 1.449 0.420 0.528 0.225 1.233 1.447 1.529 3.285 1.537 2.836 60.6 %
Public square 1.629 1.257 1.506 1.098 1.823 1.383 1.363 2.067 1.929 2.232 43.8 %
Shopping mall 1.026 0.709 0.982 0.760 1.228 0.994 1.023 1.646 0.883 1.009 63.6 %
Street, pedestrian 1.790 1.185 1.557 1.841 1.719 1.321 1.771 2.517 2.402 1.801 30.60 %
Street, traffic 1.431 0.908 1.540 1.011 1.104 1.271 1.654 1.957 1.173 2.262 64.0 %
Tram 1.680 1.358 1.570 1.207 1.056 1.765 1.298 1.840 2.714 2.303 42.2 %
Average 1.551
(± 0.036)
0.998 1.447 1.220 1.611 1.525 1.454 2.017 1.705 1.982 46.4 %
(± 0.9)

As discussed here, devices S4-S6 are used only for testing not for training the system.

Note: The reported baseline system performance is not exactly reproducible due to varying setups. However, you should be able obtain very similar results.

Subtask B

The baseline system for Subtask B is based on OpenL3 using both audio and video branches. The audio and video embeddings are extracted according to the original OpenL3 publication, after which each branch is trained additionally for the scene classification task using a single modality. The trained subnetworks (audio and video subnetworks) are then connected using two fully-connected feed-forward layers of size 128 and 10.

Repository


Parameters

  • Audio embeddings using L3 network:
    • Content type: “env”
    • Input representation: “mel256”
    • Embedding size: “512”
    • Hop size: 0.1
  • Video embeddings using L3 network:
    • Content type: “env”
    • Input representation: “mel256”
    • Embedding size: “512”
  • Audio sub-networks:
    • Input shape: 1 * 512 (embedding of 1 seconds)
    • Architecture:
      • Dense layer #1: Dense layer (512) + Batch normalization + ReLu activation + Dropout (rate: 20%)
      • Dense layer #2: Dense layer (128) + Batch normalization + ReLu activation + Dropout (rate: 20%)
      • Dense layer #3: Dense layer (64) + Batch normalization + ReLu activation + Dropout (rate: 20%)
      • Dense layer #4 (output layer): Dense layer (10)
    • Optimizer: Adam (learning rate 0.0001,weight_decay=0.0001)
    • Learning: 200 epochs (batch size 64), data shuffling between epochs
    • Loss: cross entropy loss
  • Video sub-networks:
    • Exactly same as audio subnetwork, except the input is the video embeddings
  • Early Fusion Audio-Visual networks:
    • Input shape: 1 * 1024 (concatenate audio and video embeddings)
    • Architecture: Same as audio and video subnetworks described above
    • Optimizer: Adam (learning rate 0.0001,weight_decay=0.0001)
    • Learning: 200 epochs (batch size 64), data shuffling between epochs
    • Loss: cross entropy loss
  • Audio-Visual networks (baseline):
    • Require pretrained weights
    • Audio pretrained weights (get from audio subnetwork)
    • Video pretrained weights (get from video subnetwork)
    • Input shape: 1 * 512 (concatenate the output of linear layer #2 from audio and video subnetworks)
    • Architecture:
      • Dense layer #1: Dense layer (128)
      • Dense layer #2 (output layer): Dense layer (10)
    • Optimizer: Adam (learning rate 0.0001,weight_decay=0.0001)
    • Learning: 200 epochs (batch size 64), data shuffling between epochs
    • Loss: cross entropy loss

Model selection: - Approximately 10% of the original training data is assigned to the validation set, split done such that training and validation sets do not have segments from the same location. - Model performance after each epoch is evaluated on the validation set, and best performing model is selected

Results for the development dataset

Baseline
(audio-visual)
Audio subnetwork Video subnetwork Early fusion
(audio-visual)
Scene class Log loss Accuracy Log loss Acc Log loss Acc Log loss Acc
Airport 0.963 66.8% 0.977 66.9% 2.450 54.0% 2.117 56.5%
Bus 0.396 85.9% 0.628 78.0% 0.563 85.7% 0.284 91.8%
Metro 0.541 80.4% 1.106 60.7% 1.124 72.8% 0.461 87.7%
Metro station 0.565 80.8% 1.316 58.0% 0.495 85.2% 0.319 90.3%
Park 0.710 77.2% 0.960 73.5% 1.859 73.5% 0.705 83.2%
Public square 0.732 71.1% 1.284 54.3% 1.606 61.2% 1.073 70.6%
Shopping mall 0.839 72.6% 1.384 54.9% 2.454 45.4% 1.097 77.2%
Street pedestrian 0.877 72.7% 1.285 57.4% 1.921 58.1% 1.557 64.5%
Street traffic 0.296 89.6% 0.516 84.7% 1.336 70.7% 0.324 90.8%
Tram 0.659 73.1% 1.026 62.9% 2.677 42.8% 1.697 62.1%
Average 0.658 77.0% 1.048 65.1% 1.648 64.9% 0.963 77.4%

Citation

If you are using the audio dataset for subtask A, please cite the following paper:

Publication

Toni Heittola, Annamaria Mesaros, and Tuomas Virtanen. Acoustic scene classification in dcase 2020 challenge: generalization across devices and low complexity solutions. In Proceedings of the Detection and Classification of Acoustic Scenes and Events 2020 Workshop (DCASE2020). 2020. Submitted. URL: https://arxiv.org/abs/2005.14623.

PDF

Acoustic scene classification in DCASE 2020 Challenge: generalization across devices and low complexity solutions

Abstract

This paper presents the details of Task 1: Acoustic Scene Classification in the DCASE 2020 Challenge. The task consists of two subtasks: classification of data from multiple devices, requiring good generalization properties, and classification using low-complexity solutions. Here we describe the datasets and baseline systems. After the challenge submission deadline, challenge results and analysis of the submissions will be added.

PDF


If you are using the audio-visual dataset or baseline code for subtask B, please cite the following paper:

Publication

Shanshan Wang, Annamaria Mesaros, Toni Heittola, and Tuomas Virtanen. A curated dataset of urban scenes for audio-visual scene analysis. In 2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2021. accepted. URL: https://arxiv.org/abs/2011.00030.

PDF

A Curated Dataset of Urban Scenes for Audio-Visual Scene Analysis

Abstract

This paper introduces a curated dataset of urban scenes for audio-visual scene analysis which consists of carefully selected and recorded material. The data was recorded in multiple European cities, using the same equipment, in multiple locations for each scene, and is openly available. We also present a case study for audio-visual scene recognition and show that joint modeling of audio and visual modalities brings significant performance gain compared to state of the art uni-modal systems. Our approach obtained an 84.4% accuracy compared to 76.8% for the audio-only and 70.0% for the video-only equivalent systems.

Keywords

Audio-visual data, Scene analysis, Acous-tic scene, Pattern recognition, Transfer learning

PDF