Text-to-Speech
AI Voice Likeness
A personal AI-voice creation tool designed with safety and security at its core.
My team was asked to create a way for users to create an AI voice-likeness they could use in text-to-speech replies.
From the start, we set out to understand and decisively prevent any potential harmful uses of AI-generated voices.
My team and I exhaustively listed out all possible harmful outcomes of an AI voice generator. We grouped these outcomes into 5 broad categories: Misrepresentation, Harassment, Fraud, Privacy Invasion, Misinformation.
Although initially daunting, we quickly realized that many categories of harm—harassment, fraud, privacy invasion, and misinformation—stem from the core issue of misrepresentation. By addressing misrepresentation, we would effectively mitigate all related risks.
The end product featured a setup screen which combined into a single step the collection of an example of someone's voice as well as verification that the voice being heard was being spoken by the owner of that voice (and not simply a recording taken from somewhere else).
