Sound of COVID-19

---

Author

• The Duy Pham" dpham22@my.bcit.ca
• Matthew Harrison" mharrison62@my.bcit.ca

Extracting data

You can pull the data directly from https://github.com/iiscleap/Coswara-Data, or run the bash script extract_data.sh under the data folder.

This file will auto pull the sound repository, then extract all the sounds into the data folder.

Data pipeline

Basic pipeline:

.
├── README.md
├── data
│   ├── Coswara_Data
│   ├── extract_data.sh
│   └── extracted
├── requirement.txt
└── src
    ├── data_config.py
    ├── export_data.py
    ├── test_model.py
    ├── train_model.py
    └── utils.py

• data_config.py: This file contains all the configurations for the pipeline.

Flag	Type	Details
EXPORT_IMAGE	boolean	to export spectrogram and mel spectrum image
APPLY_MFCC	boolean	to apply MFCCs on the spectrogram data
PREFIX_INPUT	string	the prefix folder on where you place the source Coswara repository
EXTRACTED_DATA_PATH	string	the path where the data from Coswara project is extracted
PREFIX_MODEL	string	the path where model will be saved
PREFIX_OUTPUT	string	the prefix folder for the exported data. This export data is the Mel Spectrogram data and needs to be export to a folder for subsequent retraining.
POSTFIX_MODEL	string	the postfix name for model
BANNED_ID	list	some records get corrupted, and so their ID will be placed here
BANNED_ID_BY_FEAT	dict	some records get corrupted at a specific feature, and they will be filtered out for that particular feature
SEED	string	seed value for randomize
index_col	list	which column is the label index - aka y
key_col	list	which column is the training index - aka X

• export_data.py: run once to generate the MFCC data. Note that the Output folder defines under of data_config

• train_model.py: Once export_data.py has been executed, this script to train our model, using the generated Mel Spectrum output

  • The model will then exports into the model folder (PREFIX_MODEL) with postfix name POSTFIX_MODEL

• utils.py: Utilities class.

• test_model.py: to test the efficiency of the model using test data

Step to install

1. Git clone the repository
2. Either run extract_data.sh, or manually download Coswara_Data and extract the data somewhere
3. Edit data_config.py to match the input path and output path.
4. Run export_data.py to generate the Mel Spectrum data
5. Run train_model.py to start training our first model
6. Run test_model.py to test result of the trained model

Packages

• Everything should be listed under requirements.txt