Tree Cutting Sound Detection Using Deep Learning Techniques Based on Mel Spectrogram and MFCC Features

This paper presents an approach that classifies the process of tree cutting events in a forest based on sounds detection. The forest environment consists of different types of sounds generated with different frequencies from various directions. CRNN model provides accurate results through multi-dimensional data input through which can learn various audio features for predicting the real-time scenarios. In this research, CRNN has implemented for performing sound event classification and predict the real-time scenarios in better way. Multiple features such as MFCCs and Mel Spectrogram are extracted from each sound sample for generating the prediction rate. In our research, we have taken three classes of data set related to tree cutting such as axe knocking, saw cutting, and tree falling. And the rest of the sounds is considered as noise class and is added to false ratio. The performance obtained through the proposed model has generated an accuracy of 93.4% on our newly proposed datasets which is better than existing methodologies of forest environmental sound detection.