• If you don't have it already, download and install either the Anaconda or the Miniconda Python 3 distribution.

• Create a conda environment with Python 3.5 conda create -n lostanlen_icassp2018 python=3.5

• Activate environment lostanlen_spl2017 source activate lostanlen_icassp2018

• Install Tensorflow (Apache License 2.0) for deep learning conda install tensorflow

• Install Keras (François Chollet, MIT License) for deep learning conda install keras

• Install pandas (BSD 3-Clause license) for parsing annotations conda install pandas

• Install pysoundfile for audio I/O conda install -c carlthome pysoundfile

• Install muda (Brian McFee, ISC license) for data augmentation pip install muda

• Install Vesper (Harold Mills, MIT license) for Old Bird flight call detector conda install -c haroldmills vesper

• Install mir_eval (Colin Raffel, MIT license) for evaluation conda install -c carlthome mir_eval

• Install pescador (ISC License) for stochastic multi-stream sampling conda install -c conda-forge pescador

• Install scikit-learn (BSD License) for support vector machines conda install -c anaconda scikit-learn

• Install skm (Justin Salamon) for spherical k-means pip install git+git://github.com/justinsalamon/skm

• Results of the ICASSP 2018 paper by Lostanlen et al. are reproducible by running Python scripts 001 to 025 in this order, from the src folder. Arguments to these scripts are provided in the command line. The folders sbatch/ID/sbatch contains files for scheduling jobs with a Slurm workload manager, like the one that is maintained in the high-performance computing facility at NYU. These scripts can be generated programmatically by running Python scripts sbatch/generate_ID.py. The log files produced by the job manager are written to sbatch/slurm. We make them available to the general public, as they contain information on running times and memory footprints for each stage of the benchmark.