Experimenting with Professional Microphones to Apply Acoustic Event Detection to Unmanned Aerial Vehicles

Unmanned aerial vehicles (UAVs) are already in use for a wide variety of tasks for which they are equipped with various sensors such as video and thermal imaging cameras or flight assistance systems. Acoustic sensors on the other hand have not yet been widely adopted for UAVs, even though they offer a wide range of possible applications in combination with AI-based signal processing. Acoustic event detection (AED), e.g., could extend the sensing capabilities of UAVs by enabling them to react to user-defined acoustic events automatically. Possible scenarios include locating emergency situations, autonomous flight to the event location, automatic monitoring of the acoustic environment, or triggering an alarm. The acoustic sensor system required for AED must be robust to external environmental conditions as it is inevitably subject to perturbations such as the air flows through the rotors, operational noise, or wind noise. In this work, we have investigated a set of commercially available microphones for their suitability for such a system. The H520 hexacopter by Yuneec International Co. Ltd. was used as the carrier. We applied linear support vector machines (SVMs) in conjunction with OpenL3 embeddings for the realization of AED. To train the system, a dataset from the DCASE2020 challenge has been extended with recorded noise to consider different use cases.