Recurrent Neural Networks for Environmental Sound Recognition using Scikit-learn and Tensorflow
Chinnavat Jatturas, Sornsawan Chokkoedsakul, Pisitpong Devahasting Na Ayudhya, Sukit Pankaew, Cherdkul Sopavanit, Widhyakorn Asdornwised · 2019
In this paper, we perform comparison techniques for environmental sound classification with multilayer perceptron (MLP) and support vector machine (SVM), and deep learning using new machine learning platforms, i.e., Scikit-learn and Tensorflow, respectively. For feature-based classification, principal component analysis of short-time Fourier transform is used as our feature as the front end to MLP and SVM. For deep learning-based classification, convolution+pooling layers is acting as feature extractor from the input image, while fully connected layer will act as a classifier. Our experimental results show that our proposed deep neural network (DNN) models outperform the feature-based sound classification algorithms and the original deep learning work [1].