Voice Auth: Audio Verification with Deep-Learning

Today, AI has been helping us in every aspect, from our phones to driving cars on itself. However, AI can also prove to be a boon to cybercriminals as they also evolve their attack techniques. Using AI to generate fake audio, unrecognizable from the real one has been prevailing to deceive the current voice authentication system. It also becomes difficult to perform audio forensics on these files and might go unnoticed. The given system manipulates the various audio features extracted from the audio such as spectrogram, chromagram, and Mel-spectrum, etc. And differentiate between the real one and the deep-faked voice. This review paper discusses the two for the detection of fake audio. The extracted feature is visualized by converting it into a PNG. Employing CNN to differentiate between the real and deep-faked voice and adjusting the +network appropriately gives the best results. The system efficiently reduces the effort of audio forensics and gives accurate results. RVC-based cloning can be detected using various audio features. The system can also be used for voice authentication where deep fake voices can deceive the existing system.