Uttera
Transcribe
Audio to text, with the language detected on its own.
POST /v1/audio/transcriptions
Transcribe, summarize, translate and speak. Over an API or from your browser, with nothing to install. OpenAI-compatible: if you already use it, you change two lines.
Try it right here Read the docs
Processed entirely in Spain · We don't keep your audio · We don't train on your data · Open source
Complies with the GDPR and Spain's LOPDGDD
Five services over the same audio. Billed by the second, not by the request.
Audio to text, with the language detected on its own.
POST /v1/audio/transcriptionsText to audio, from a voice catalog or from a voice you supply.
POST /v1/audio/speechA recording in one language; text and voice in another.
POST /v1/translateTone, speaker profile, and who is talking at each moment.
?extras=sentiment,profile,diarizeA whole recording, as a structured summary.
POST /v1/summarizeThe voice is delivered as it is generated: the first sound comes out immediately.
POST /v1/audio/speech/streamSame routes, same parameters, same voice names. The official SDK works without patching anything: you change the address and the key.
And with one difference that decides it for a lot of companies: the audio never leaves the European Union.
from openai import OpenAI
c = OpenAI(
api_key="sk-echo-…",
base_url="https://api.uttera.ai/v1",
)
audio = c.audio.speech.create(
model="tts-1", voice="nova",
input="Your order ships tomorrow.",
)Nothing else. Uttera is not an archive.
Your data stays yours, without exception.
Audio is processed in memory and discarded when we answer. The transcript, the analysis and the summary travel in the response and are not stored. The only thing written to disk — the audio we generate, for one hour — you can turn off yourself on every request.
Our own data centers in southern Spain. No third-party clouds and no international transfers to justify.
The models — including the one that summarizes — run on our own hardware. There are no calls to OpenAI, or Google, or anyone.
Your audio and your text train no model. They don't become statistics either, or "anonymized data", or study material. There is no fine print behind that sentence: no copy survives to do it with. The models learn nothing from what passes through them: every request begins and ends without leaving a trace. If we ever train one of our own it will be on data licensed for that, never on yours.
Apache 2.0, and they are the same ones running in production. You can read and audit our code. If your case can't have the audio leaving your network, run it yourself.
A minute of audio or a paragraph of text. No account, no card.
The demo uses the standard voice and is rate-limited. A free account gets you the whole catalog, voice analysis and summaries.
An account, a key and credits to try it. No card and no payments.