VoiceMorph: How AI Voice Morphing Reveals the Boundaries of Auditory Self-Recognition

Most people lose recognition of their own voice at around 35% morphing, meaning AI can alter a voice enough that the original speaker no longer identifies it as their own—raising concerns about voice-based authentication and identity verification systems.

Older adults are significantly more vulnerable to voice morphing, tolerating much higher levels of alteration before noticing the change (β = 0.617, p = 0.048), which suggests this demographic may face heightened risks from voice-based fraud or impersonation as voice synthesis technology becomes more accessible.

People use prosodic patterns (rhythm, intonation, and stress) rather than just pitch to recognize their own voices, but this strategy can be manipulated by AI, indicating that current voice-based security measures may have fundamental limitations that need to be addressed.

Participants expressed universal discomfort about voice manipulation technology, even while acknowledging legitimate uses like assistive technology, highlighting a gap between the potential benefits and public concern about misuse—suggesting the need for clear ethical guidelines and regulations before widespread deployment.

Response times slow dramatically when people encounter highly altered or cloned versions of their own voices, suggesting that voice morphing creates cognitive confusion that could be exploited, underscoring the importance of developing detection tools and protective measures for vulnerable populations.