deep-learning

project Urban Sound Classification Audio classification is an important area of interest with multiple applications in medical, industrial and multimedia domain. Sound or audio is also considered to be the integral part of security and video based surveillance systems. Sound can be represented as raw wave form and spectrogram. Different approaches are being used to build and train neural networks like convolutional neural networks and recurrent neural networks for the identification of Urbansounds using raw wave forms or spectrograms. The purpose of this study is to classify sound using two approaches , one on the basis of raw audio by a convolutional neural network and second on the basis of spectrogrms by transfer learning on GoogLNet. The purpose of using GoogLeNet is to take advantage of a well known standard model with its pretrained weights. Performance of these two approaches comapred on the basis of accuracy, precision and re-call. Urbansound8k is a well know dataset that is used for audio and image classification.