Cough detection with Log Mel Spectrogram, Wavelet Transform, Deep learning and Transfer learning concepts.
• A computer vision and deep learning (with transfer learning) framework for detection of cough sounds in the presence of other common household sounds. • Log Mel spectrogram and Wavelet transform images were obtained for sound data samples (ESC-50) • A Convolutional Neural Networks based pretrained model "VGG16" was trained on this image dataset to detect cough sounds with high test accuracy.
More details about the ESC-50 data: https://github.com/karolpiczak/ESC-50
More details about VGG16 model - https://neurohive.io/en/popular-networks/vgg16/
More details about Log Mel Spectrograms - https://towardsdatascience.com/getting-to-know-the-mel-spectrogram-31bca3e2d9d0
For more cough sound recordings data, check out IISc CoSwara project at https://github.com/iiscleap/Coswara-Data