How to hear an AI voice
Breath, room tone, and uneven emphasis are the practical checks for synthetic speech.
Play the clip once for the words and a second time for the sound around the words. Synthetic speech is often fluent and slightly too even. Real speech wastes air.
Listen for a breath before a long sentence, a tiny mouth noise, and a room. A voice with no space around it, as if it were recorded in nowhere, is a reason to be suspicious. So is a voice whose volume never dips, even on the unimportant words.
Prosody is the other half. People stress the unexpected word, trail off, and restart. A model reading a paragraph tends to give every sentence the same contour, like a careful announcer who never gets tired.
None of this catches a short, well-produced clip by itself. A real booth recording is also dry and even. Use the checks together, and use the voice quiz where the label stays hidden until you commit.