An Improved Acoustic Scene Classification Method Using Convolutional Neural Networks (CNNs)
Predicting acoustic environment by analyzing and classifying sound recording of the scene is an emerging research area. This paper presents and compares different acoustic scene classification (ASC) methods to differentiate between different acoustic environments. In particular, two deep learning techniques of classifica-tion i.e. Deep Neural Network (DNN) and Convolution Neural Network (CNN) have been applied using a combination of Mel-Frequency Cepstral Coefficients (MFCCs) and Log Mel energies as features. DNN and CNN are state-of-the-art techniques which are being used widely in speech recognition, computer vision, and natural language processing applications. These techniques have recently achieved great success in the field of audio classification for various applications. Both techniques have been implemented and tuned by performing a variety of experiments with different hyper parameters, hidden layers and units on public benchmark datasets provided in the DCASE 2017 challenge. The proposed method uses frame level randomization of the combined acoustic features i.e. MFCC and log mel energy, for training of model to achieve higher accuracy with DNN and CNN. It has reported higher accuracy than the previous work done on public benchmark datasets provided in the DCASE 2017 challenge. It is observed that DNN achieved 83.45% and CNN achieved 83.65% accuracy that is higher than the previous work done on public benchmark datasets provided in the DCASE 2017 challenge.
- There are currently no refbacks.