What Uttera is

Uttera works with voice. You give it audio and it hands back text, analysis or more audio; you give it text and it hands back speech.

It is built for what happens after a recorded conversation — a call, a meeting, an interview, a voice note: transcribing it, knowing who spoke and in what tone, translating it or summarizing it. It is not a text translator or a document reader: audio is at one end or the other, always.

Two ways to use it

You don't need to know how to program to use Uttera. There are two doors, and most people only need the first:

Both do exactly the same thing and cost exactly the same: the Studio is not a cut-down version or a demo, it's the same machine with a button in front. The only difference is who presses the button — a person or a program.

What is an API?

API stands for Application Programming Interface. Without the jargon: it is the door through which one program asks another program for something.

The comparison that works best is a restaurant. You don't walk into the kitchen: you tell the waiter what you want, they take it inside and come back with the dish. You don't need to know how the kitchen is laid out, which oven they use, or how many cooks there are. You only need to know what you can order and how to order it. The API is the waiter, and this documentation is the menu.

In practice, a program of yours sends an audio file over the internet to an address of ours, along with a key that says who you are, and gets the result back. Nothing is installed on your computer: the work happens on our machines and only the request and the response travel.

The name

Uttera /ˈʌt.ər.ə/ comes from the English verb to utter: to pronounce, to say aloud, to give audible expression to something.

Formally it is also a backronym of Universal Text Transformer Engine for Realtime Audio, which captures its origin — a speech-to-text and text-to-speech server — and the Transformer architecture it is built on.

What it can do

Concepts

Credits

Everything is billed in credits. One credit is one second of GPU time, and each service has its coefficient according to what it actually consumes. There are no per-request charges and no minimums: you pay by second of audio processed or generated.

Your plan gives you a monthly allowance of credits that renews on your subscription date, not on the 1st of the month. What you don't spend does not carry over.

Voices

There are two families. The standard voices are a fixed catalog, fast and cheap. The cloned voice reproduces a particular timbre from a sample, costs about 19 times more per second, and is not in every plan.

Duration, not size

Billing goes by seconds of audio, not by megabytes. The same minute of speech costs the same in WAV as in MP3, even though the file is twenty times heavier.

Your first request

You need a key. You create it in your account; it starts with sk-echo- and is shown only once.

curl -X POST https://api.uttera.ai/v1/audio/transcriptions \
  -H "Authorization: Bearer $UTTERA_API_KEY" \
  -F file=@recording.mp3 \
  -F model=whisper-1 \
  -F response_format=text

Response:

Good morning, I was calling about last week's order...