An MFCC-based Secure Framework for Voice Assistant Systems

Voice assistant systems accept voice commands to perform various routine tasks such as navigation, playing music, temperature control, and so on. Voice assistants are commonly employed in home automation to serve the elderly and disabled people. Despite of their popularity in our daily lives, voice assistants are still far from perfect as these systems suffer from lack of security and accessibility issues. Available voice assistants respond to voice commands from any individual without proper user authentication. Hence, these systems are vulnerable to various security attacks where any adversary can control the voice assistant and perform any desired action. Moreover, popular voice assistants are also platform-dependent and only function on specific kind of devices. In the light of these limitations, we propose a Mel-frequency cepstral coefficients MFCC-based framework to secure the voice assistants via user authentication. MFCC efficiently recognizes the voice features which are then employed to distinguish between authorized and unauthorized users. Moreover, the voice assistant for the proposed framework is also implemented via cross-platform technology to guarantee increased accessibility. The proposed framework is tested on authorized and unauthorized user voices under different conditions with positive outcomes. The positive results prove the feasibility of our proposed framework for securing the voice assistant systems.