Repository for the 3A Project - Reconnaissance de l'activité de vie quotidienne humaine à domicile à l'aide du son - IMT Atlantique - 2021/2022

• Mateo BENTURA
• Ezequiel CENTOFANTI
• Kevin MICHALEWICZ
• Oumaima TOUIL

In order to be able to execute the following steps, you will need to create a Python Environement. This can be done by:

pip install -r requirements.txt

├── README.md                                   # This file
├── requirements.txt                            # Required packages
├── PythonAudioClassification.ipynb             # Classification implementation notebook, to train and test
├── AudioClassifier.py                          # Module to store defined functions and classes
├── test_audios.py                              # Simple testing script to classify a single audio file
├── .gitignore                                  
|
├── LL_AudioDB                                  # Sound database
|   ├── audio                                   # All audio files
|   ├── custom-audios                           # Our recorded audio files
│   ├──  metadata                                
|   │     ├── kitchen20b.csv                    # List of audio tracks 
|   │     └── kitchen20.csv                     # DB format
|   ├── add_audios.py                           # Script to add new audios to DB
|   └── categories.txt                          # List of categories used by 'add-audios.py'
|
├── Old-k20-model                               # First implementation of k20 model (unused)   |
└── checkpoints                                 # Trained model checkpoints

To train the model or bulk test audios we can use the notebook PythonAudioClassification.ipynb, whose use is straight-forward and requires simply running each command, with the option of loading the pre-trained model.

If you want to simply test one audio file, a testing script called test_audios.py is available. To used this, simply type the following command

python test_audios.py input_audio.wav

An exemplary output would be the following:

Testing audio 'input_audio.wav'
Predicted label is 'trash'

This implementations uses a simple database, consisting of a directory and a .csv metadata file. In order to update this database and include new audio files, we provide a Python script called add_audios.py. This takes the .wav files from an input folder (they must indicate the label in the name) and places them in an ouput folder, updating the metadata file as well.

To call this script we use the following command in the corresponding directory:

python add_audios.py

This uses the default input and output directory names custom-audios and audios, we can use custom ones with the command:

python add_audios.py input-folder output-folder

The script prints the advancements and asks for a confirmation of this update.