Comparison of Mel Frequency Cepstral Coefficient (MFCC) and Mel Spectrogram Techniques to Classify Industrial Machine Sound

The utilization of sound recognition to detect the faulty machines are getting the interest to be explored with the technology advancement in IR 4.0. In this paper, we investigated the performance of two acoustic features in representing the sound of a fan in industrial machines, that is acquired from the MIMII dataset. The Mel Frequency Cepstral Coefficient (MFCC) and Mel Spectrogram techniques are explored and compared in this paper. The classification is performed using Convolutional Neural Network (CNN) with the input of images from both techniques. The dataset of the input images are the audio files of 6dB fans sound in industrial machine. The results suggest that Mel Spectrogram is an effective feature for faulty machine sound recognition when used with CNN model.