Audio Classification for Noise Filtering Using Convolutional Neural Network Approach

Abstract

In each part of daily routine, sound assumes a significant part. From discrete security features to basic reconnaissance, a sound is a vivacious component to create automated frameworks for these fields. Scarcely any frameworks are now on the lookout, yet their effectiveness is a concerned point for their execution, real-time conditions. The learning capacities of Deep learning designs can be utilized to create sound characterization frameworks increase the impact of sound classification. Our main aim in this paper is to implement deep learning networks for filtering the nose and arrangement of these sound created by the natural phenomenon’s according to the spectrograms that are created accordingly. The spectrograms of these natural sounds are utilized for the preparation of the Convolutional neural network (CNN) and Tensor Deep Stacking Network (TDSN). The utilized datasets for analysis and creation of the networks are ESC-10 and ESC-50. These frameworks produced from these datasets were efficient in accomplishment of filtering the audio and recognizing the audio of the natural sound. The precision obtained from the developed system is 80% for CNN and 70% for TDSN. Form the implemented framework, it is presumed that proposed approach for sound filtering and recognition through the utility spectrogram of their subsequent sounds can be productively used to create efficient frameworks for audio classification and recognition based on neural networks.