Uttera · Use it where you already use OpenAI

The two lines

This is literally everything that changes.

Before

from openai import OpenAI

client = OpenAI(
    api_key="sk-proj-…"
)

After

from openai import OpenAI

client = OpenAI(
    api_key="sk-echo-…",
    base_url="https://api.uttera.ai/v1"
)

The rest of the file is untouched.

What works the same

Checked against the official SDK, not inferred from the docs.

Text to speech

client.audio.speech.create(
    model="tts-1",
    voice="dora",
    input="Hola.")

With tts-1 and tts-1-hd, as there. The voices are ours: GET /v1/audio/voices lists them.

Transcribe

client.audio.transcriptions.create(
    model="whisper-1",
    file=f)

Whisper large-v3-turbo underneath. Takes wav mp3 flac ogg opus aiff m4a webm.

List models

client.models.list()

Returns tts-1, tts-1-hd and whisper-1. Handy to check at a glance that the key and the URL are right.

What doesn't, and we say so here

Promising full compatibility and letting you crash in production would be worse than promising nothing.

What you will not find

chat.completions, embeddings, images and the rest return 404. Uttera does audio: it is not a replacement for all of OpenAI, it is a replacement for its audio part.

If your application uses both, keep the OpenAI client for text and create a second client for audio. Four lines.

What there is on top

Outside the OpenAI shape there are things that do not exist there: translating a recording into another language with voice, a structured summary, who speaks and when, tone and speaker profile, sound effects and music.

They are called the same way, with the same client and the same key. They are all here →

And why you would switch

Because if it were only «the same but from someone else», there would be no reason.

It does not leave Spain

Audio is processed on our own hardware in Spain. There is no international transfer to justify, no standard clauses, and no question about which law reaches your data. Why that decides purchases →

We don't train on your data

Not yours and not anyone's. Audio is processed and discarded on answering: no copy is left to do it with, even if we wanted to.

It can be read

The engines are published under an open licence. A privacy claim you can only take on faith is worth less than one you can read. The code →

Check it in a minute

With nothing installed beyond what you already have.

export UTTERA_API_KEY=sk-echo-…

python3 - <<'PY'
import os
from openai import OpenAI
c = OpenAI(api_key=os.environ["UTTERA_API_KEY"],
           base_url="https://api.uttera.ai/v1")
print([m.id for m in c.models.list()])
r = c.audio.speech.create(model="tts-1", voice="dora", input="It works.")
open("prueba.mp3", "wb").write(r.content)
PY

There are equivalent examples in curl and Node in uttera-examples →

Start free

An account, a key and credits to try it. No card.

Create account See the documentation