A Study on the Sound Recognition Method of Autonomous Vehicle using CNN

In this paper, a study on the algorithm that recognizes and judges sound source using convolutional neural network (CNN) is introduced. It is assumed that multiple of microphones are attached to receive sound information. The received sound information is then converted to visual information with the Mel-spectrogram which expands 1-dimensional sound information to 2-dimensional information. However, the shorter the extraction time by reducing n_mels, the lower the resolution of the image and the lower the performance as learning data. The value of n_mels = 64 is suggested to minimize the extraction time of Mel-spectrogram because this algorithm should be used in the autonomous vehicle. Through the computational experiment, 95% accuracy was obtained through CNN, machine learning.