Robust Deep Learning Framework for Real-Time Denoising of Heart Sound

Requirements

• Python
• Matlab
• Keras
• Tensorflow
• Sklearn
• Tensorboard
• naturalsort
• keras-flops
• librosa
• soundfile
• numpy
• pandas
• matplotlib

How To Run

Training:

• First download the PHS Data (Processed) and ICBHI Dataset (Processed) folder from GoogleDrive Link provide inside Data/data_download_link.txt file
• Update the definition of path_Heart_Train and path_Lung_Train and specify the model name (for example: use 'lunet' for proposed denoising framework) under the Codes/config.py file
• Run Codes/train_model.py to start the training

Re-Generate Results:

Open-access Heart Sound Dataset

• First download the OAHS Dataset, Hospital Ambient Noise (HAN) Dataset, and ICBHI Dataset (Processed) folders from GoogleDrive Link provide inside Data/data_download_link.txt file
• Update the definition of pathheartVal, pathlungval and pathhospitalval under the Codes/config.py file
• Put the directory of training weight(you can find pretrained weight inside Models folder) inside Codes/result_making.py file
• Run Codes/result_making.py to start the inference

PASCAL Heart Sound Challenge Dataset

• First download the PaHS Dataset provided inside Data/data_download_link.txt file
• Update the definition of pathheartVal, pathlungval and pathhospitalval under the Codes/config.py file
• Put the directory of training weight (you can find pretrained weight inside Models folder) inside Codes/result_making.py file
• Run Codes/result_making.py to start the inference
• Use the directory of the generated .csv file (containing the denoised audio samples) inside the readtable function of Run Codes/SNR Estimation Algorithm/SNR_Estimation_Denoised.m to get the estimated SNRs for the denoised signals
• Run Codes/SNR Estimation Algorithm/SNR_Estimation_Noisy.m to get the estimated SNRs for the noisy signals