Audio Scene Classification Based on Convolutional Neural Networks: An Evaluation of Multiple Features and Topologies in Short Time Segments

The classification of acoustic scenes has been gaining importance in recent years, either because of the applications that it may have or because of the challenge of implementing a computational tool that allows the proper detection of complex and diverse sounds, such as those presented in real environments. In this work a convolutional neuronal network is implemented, trained with features such as mel-frequency cepstral coefficients, gamma tone y discrete Fourier transform, extracted to the sounds in segments of 10 s and 1 s. Cross validation is used with 80% of the data for training and 20% for validation with the DCASE2018 database evaluating the performance for different topologies of the neural network.