InstallationΒΆ
Download the system
Download the latest package or clone the system directly from the repository. To clone repository with HTTPS:
git clone https://github.com/TUT-ARG/DCASE2017-baseline-system.git
or with ssh:
git clone git@github.com:TUT-ARG/DCASE2017-baseline-system.git
The system is developed for Python 2.7, Python 3.5 and Python 3.6. The system is tested to work on Linux, Windows and MacOS platforms.
One can install either the official CPython or use some Python distribution based on it. New users are recommended to use Anaconda Python distribution.
Using the system on Windows
The baseline system uses rather long directory paths, since it is storing 32 character MD5 hash of the system parameters into directory names. Some Windows system have path length limit (lowest 260 characters), which is causing problems. To avoid problems related to this, install the system as close as possible to the drive root.
External modules
To ensure that all external modules are installed, run command:
pip install -r requirements.txt
PySoundFile
PySoundFile is used to read and write audio files in the system. The library depends on system level library libsndfile. Under Linux (Debian/Ubuntu) you can install this with command:
sudo apt-get install libsndfile1
Theano
The system uses by default Theano as Keras backend.
There was a bug in Theano 0.8.2 version, so make sure to use 0.9.0 release. The bug will affect Task 2 system (when using GPU and binary_crossentropy as loss function). To fix this, make sure you have installed correct version of Theano:
pip install theano==0.9.0
Or use latest from git with command:
pip install --upgrade --no-deps git+git://github.com/Theano/Theano.git
Tensorflow
It is also possible to use Tensorflow as Keras backend. If you plan to use it with GPU make sure you have installed GPU enabled package:
pip uninstall tensorflow
pip uninstall tensorflow-gpu
pip install tensorflow-gpu
Current version of Tensorflow only supports CUDA 3.0 compatible graphic cards.
Audio datasets
The system will automatically download the needed audio datasets, and place them under the directory specified in the parameters (see parameter path->data).
| Dataset | Type | Audio files | Size on disk | License |
|---|---|---|---|---|
| TUT Acoustic scenes 2017, development | Acoustic scene | 4680 | 22Gb | Academic use only (see EULA inside the package) |
| TUT Rare sound events 2017, development | Sound events / synthetic | 1281 | 9.2Gb | Academic use only (see EULA inside the package) |
| TUT Sound events 2017, development | Sound events realistic | 24 | 2.6Gb | Academic use only (see EULA inside the package) |