Play a modern text-to-speech clip to someone and they will tell you it sounds human. Put the same voice on a phone call and they will know within ten seconds that it is a machine. The voice was never the problem.
What gives it away is everything around the voice. A real call has a room behind it. It has the sound of somebody moving in a chair, a keyboard, another person on another call. And when you have to wait, something happens: you hear music, or you hear the person say “bear with me”. What you never hear, on a call with a human being, is perfect, absolute silence.
The vacuum is the tell
An AI agent thinking is dead air. It is a specific, unnatural kind of nothing: no line noise, no breath, no room. Humans are extraordinarily good at detecting it, because a silent line has only ever meant one of two things — the call dropped, or nobody is there.
- A caller who hears a beat of pure silence assumes the line is dead and starts saying “hello? hello?”
- By the time the agent answers, the caller has already decided something is wrong.
- Every subsequent perfect sentence now reads as a machine, because the frame is set.
What we did about it
Two things, and they are the least glamorous features we have ever built. First, a room: you can put a background behind your agent, from a busy call centre to a quiet office to a clinic waiting room. It sits far below the voice, and a caller only notices it if it is missing.
Second, we cover the real waits. When a specialist is joining the call, or a colleague's phone is ringing, the caller hears music instead of nothing. But — and this is the part everyone gets wrong — a short pause stays silent. Cutting to music because someone paused for a second is just as unnatural as the silence was. Below two and a half seconds, nothing plays.
You can hear it, which is the point
Every voice-AI company on earth claims to sound human, and no buyer has ever been able to check. So we made ours playable: the rooms and the tunes your callers would hear are on the voices page, in your browser, with no signup. Press play and judge it.
It is worth being honest about why this matters commercially, too. You do not lose a deal because your AI mispronounced a word. You lose it because the customer worked out it was an AI and stopped taking the conversation seriously.
You can build an agent free, hear one live, or see the pricing before you commit to anything.