Deep learning

Physics

Sound recording and reproduction

Pattern recognition (psychology)

Spectrogram

Speech coding

Audio signal processing

Mel-frequency cepstrum

Artificial intelligence

Acoustics

Computer science

Feature extraction

Audio signal

Speech recognition

Artificial neural network

Convolutional neural network

This document illustrates a simple audio classification model based on deep learning. We address the problem of classifying the type of sound based on short audio signals and their generated spectrograms, from labeled sounds belonging to 10 different classes during model training. In order to meet this challenge, we use a model based on Convolutional Neural Network (CNN). The audio was processed with Mel-frequency Cepstral Coefficients (MFCC) into what's commonly known as Mel spectrograms, and hence, was transformed into an image. Our final CNN model achieved 91% accuracy on the testing dataset.

Urban Sound Classification using CNN