Voice Signature Recognition for UAV Pilots Identity Verification

Employing deep learning as a soft computing tool encouraged researchers to design intelligent systems that mimic human behavior. Speaker identification is a computing task to validate the user claimed identity using features extracted from his voices. These unique features allow researchers to identify a human's voice using different machine learning techniques that are proved to be efficient in the areas of voice recognition, especially for security purposes. Unmanned Aerial Vehicles (UAVs) have gained noticeable attention in recent years for a variety of applications. Simplifying the UAVs control system by adding a voice recognition add-on will allow larger segment of people to enjoy and benefit from such a practical technology. However, a designer should take care of the security and privacy of the user and make sure that the voice recognition system responds only to the voice of the owner. Therefore, we propose, in this paper, a speaker identity verification add-on for security and privacy of voice-controlled UAVs using a small size dataset. The speaker identity verification was implemented in two stages: extracting features from short time recorded audio clip by computing Mel-Frequency Cepstral Coefficients (MFCCs) and applying a feature matching. The results showed a high identity verification accuracy, which makes the proposed approach is a practical add-on to be used for the security purposes of voice-controlled UAVs.