Do Synthetic Voices Have Emotions? Exploring Emotional Cues for Spoofed Speech Detection
Cathal Fitzgibbon, M. S. Arun Sankar, Utz Roedig · 2024
Spoofed speech poses a significant threat as it can be used to impersonate people. Spoofed speech can also be used to gain unauthorized access to systems where speech is used as a biometric identifier. It is therefore essential to develop methods that can distinguish bonafide and spoofed speech. Generated speech quality is improving dramatically in recent years and it is necessary to find new effective detection methods. In this work, we analyze the emotion of both, spoofed and bonafide speech from the ASVspoof 2019 dataset using various Speech Emotion Recognition (SER) algorithms. We show that generated (spoofed) speech lacks emotion profiles found in bonafide speech. Thus, SER algorithms can be used for spoofed speech detection and a layer of SER may be added to speech processing systems to safeguard against spoofed speech.