Drone Noise Reduction using Deep Convolutional Autoencoder for UAV Acoustic Sensor Networks

Chanjun Chun, Kwang Myung Jeon, Taewoon Kim, Wooyeol Choi

Abstract: Drones are widely utilized in various industries. Unfortunately, when a drone acquires sound through a microphone, which is installed itself, drone flying and wind noises appear in recorded signals. Therefore, it is necessary to reduce such drone flying and wind noises for enhancing the quality of the recorded sound signals for UAV acoustic sensor networks. In this paper, we proposes the noise reduction method using a deep convolutional denoising autoencoder for eliminating drone flying and wind noises. The deep convolutional denoising autoencoder is widely utilized to extract the target sound source in monaural audio source separation. To do this task, a training dataset is constructed by mixing drone flying and wind noises in clean speech signal. Also, we train the neural network model, which is in form of fully convolutional neural networks. From the sound signals recorded in the real outdoor environment, it is shown that the trained model can reduce the drone flying and wind noises, and only separate the target speech.