Welcome to International Journal of Emerging Technologies in Engineering Research (IJETER)

Volume 9, Issue 1, January (2021)

S.No	Title & Authors	Full Text
1	Transfer Learning Models Based Environment Audio Classification Jasmine Chhikara Abstract - This work proposes a transfer learning model to train Convolutional Neural Networks (CNNs) for environmental audio classification. The model uses cyclic learning rate and adam optimizer with decoupled weight decay to stabilize the training process. Audio augmentation is applied to artificially increase the size of dataset which is followed by conversion of time-series audio samples to log-scaled mel-spectrogram. The log-scaled mel-spectrogram is then fed as input to the model to provide matching performance with the baseline results. The model produces significant results with reduced training epochs on same dataset size. Different types of audio augmentations are performed and a comparative study of three classification models including Xception, MobileNetV2 and DenseNet is presented here.	Download