Turing TTS is a benchmark that pairs human and AI-generated speech of the same sentences and measures the rate at which listeners incorrectly identify the AI speech as human.