NoteFish live voice
Answer callers in their language, in your own voice, sounding the way you actually sounded.
Download for macOSIt sits next to whatever call you’re on. Give it your phone number and it picks those up as well.
Eighty-odd languages on their end. Yours on the desk.
You talk. About half a second later the caller hears you in their language, and it is still your voice. Nobody in between is relaying anything.
You, the caller, and a third person relaying. Every sentence gets said twice.
They ring from a phone, or from Zoom, or from wherever. By the end of their first sentence the desk knows which language it is and captions it in yours. It also notes how they sound, because an upset caller and a cheerful one need different answers.
Hold the key and say it the way you’d say it to the person at the next desk. Ramble a bit if you need to. When you let go, the reply is on its way.
The words alone don’t say how something was said. So the desk measures your clip against your own normal: were you louder than usual, faster, softer? It weighs that against the caller’s mood and picks the take of your voice that fits.
The caller hears your voice, in their language, carrying the feeling you had. One side talks at a time, so nobody gets talked over. And if the desk can’t confirm a reply actually played, it says so instead of guessing.
For the floor
Leave the caller’s language on Auto. They start talking, the desk hears which language it is and replies in it. Your captions stay in whatever you set as yours.
You record one short passage for each feeling: calm, warm, energetic, reassuring, apologetic, firm. During a call the desk plays whichever take matches how you just spoke. If you’d rather choose yourself, tap one.
The lines you say forty times a day. One tap says them to the caller, translated and in your voice, and your hands never leave the ticket.
One line of house style changes how every reply is worded: how long it runs, how formal it is. What it can’t change are the facts. Numbers, times and promises come out exactly as you said them.
Drag the panels around, hide the ones you never touch, save a few presets. The layout belongs to your seat, so it comes with you to whichever machine you sit at.
Seats and a queue. A call rings every free seat and the first agent to pick up gets it. Each agent has their own voice takes, canned lines and layout. Supervisors can see who is on what without hearing a second of audio.
A cloned voice is a serious thing, so we treat it like one. Nothing gets cloned without your explicit consent, and the caller’s voice is never cloned at all. Keys stay on the server. The transcript stays with the call, in your own system of record.
Learn moreFrom one real call
Sep 16, 2026 · 10:14 · French → English · Nina O.
Marie will wait at the downstairs door. The buzzer is broken; the driver should call on arrival. If it slips past 10:45, call her back.Read the call record
“Est-ce que le chauffeur va appeler avant d’arriver ?” came up on the desk as “Is the driver going to call before he arrives?” about half a second later.
The caller sounded upset, so the desk led with an apology: your apologetic take, slowed to 0.92×.
Français. Captions in English, replies in French. You didn’t set anything.
Does the caller hear a robot?
Download the Mac app and read aloud for a minute so it can learn your voice. Your next call can be in any language. That’s the whole setup.
Download for macOSRuns on the Mac next to your call: Zoom, Meet, Teams, WhatsApp, FaceTime, anything that uses a microphone. With a phone number connected it answers from any browser as well.