Usage

For each task there is a separate application (.py file) in applications/ directory:

applications/task1.py	DCASE2017 baseline for Task 1, Acoustic scene classification
applications/task2.py	DCASE2017 baseline for Task 2, Detection of rare sound events
applications/task3.py	DCASE2017 baseline for Task 3, Sound event detection in real life audio

Application arguments

All the usage arguments are shown by python task1.py -h.

-h, --help	Application help.
-o, --overwrite	Force overwrite mode.
-v, --version	Show application version.

System mode

-m {dev, challenge}

System mode, dev for normal development, challenge for challenge type behaviour

Each application has two operating modes: Development mode and Challenge mode. In development mode, the development dataset is used with the cross-validation setup: training applied for training set, and testing for testing set. In challenge mode, the development dataset is fully used for training the acoustic models, and a second dataset, evaluation dataset, is used for testing to generate system outputs (if ground truth is available for the evaluation dataset, the output is also evaluated). This mode is designed to be used when running the system on the evaluation dataset, for generating the system outputs for the challenge submission.

System parameters

-p FILE, --parameters FILE	Parameter file (YAML) to overwrite the default parameters
-s PARAMETER_SET, --parameter_set PARAMETER_SET	Parameter set id. Can be also comma separated list e.g. -s set1,set2,set3. In this case, each set is run separately.

The application supports multi-level parameter overwriting, to enable flexible switching between different system setups. The default parameters are defined in applications/parameters/task?.defaults.yaml, and these parameters are replaced by parameter set for the current run. Define here only parameters that you want to overwrite (compared to the defaults).

More about parameterization

Information

-show_set	List of available parameter sets
-show_datasets	List of available datasets
-show_parameters	Show current parameters
-show_eval	Show evaluation results

System printing

-n, --node

Node mode, console printing tuned for computer grid usage

Basic usage

With default settings, the system will download the needed datasets and extract them under directory data (storage path is controlled with parameter path->data), and proceed to train and evaluate the system, for example:

python task1.py

To run all provided system setups one after another:

python task1.py -s dcase2017,dcase2017_gpu,gmm,minimal

For development with the system, one should create a new parameter set file in order to overwrite the default parameters with it:

python task1.py -p custom.yaml -s custom_set

Example of custom.yaml file:

active_set: custom_set

sets:
    - set_id: custom_set
      feature_extractor:
        win_length_seconds: 0.1
        hop_length_seconds: 0.5

To run the system in challenge mode:

python task1.py -p custom.yaml -s custom_set -m challenge