glam-imperial / ComParE2020-Breathing-End2End

This repository contains the code to replicate the End-to-End Deep Sequence Modelling baseline for the Breathing Challenge of the Interspeech 2020 Computational Paralinguistics Challenge (ComParE).

This repository contains the code to replicate the End-to-End Deep Sequence Modelling baseline for the Breathing Challenge of the Interspeech 2020 Computational Paralinguistics Challenge (ComParE). [whitepaper]

Usage

This code was tested with Python 3.5.

Install required packages

Execute:

pip install -r requirements.txt

preferably on a separate virtualenv.

Breathing Challenge package

1. Edit end2end/configuration.py - specifically the CHALLENGE_FOLDER needs to correspond to the path in your disk the challenge folder exists.

2. Execute end2end/data_generator.py - this will process the data and extract tf_records, to be used in the experiment by tensorflow.

3. (Optional) Edit end2end/experiment/experiment_setup.py - this file defines a Python dictionary with configuration values for the experiment.

4. Execute end2end/run_experiment.py - this will perform one trial of the End2End method for the Breathing Challenge, as described in the whitepaper.

5. Adapt end2end/run_experiment.py and end2end/experiment/core.py - make your own adaptation of this method and outperform this baseline!