Legal notice
and privacy

This page brings together everything that governs the use of uttera.ai and the service behind it. It is written to be understood, not to avoid being understood.

• 1. Who operates this
• 2. Terms of use of the site
• 3. Terms of the service
• 4. Prohibited uses
• 5. Cloned voice and recordings
• 6. Privacy policy
• 7. Data processing agreement
• 8. Cookie policy
• 9. Changes and governing law

1. Who operates this

In compliance with article 10 of Spanish Law 34/2002, on information society services and electronic commerce (LSSI-CE):

2. Terms of use of the site

2.1 Purpose and acceptance

These terms govern access to and use of the uttera.ai website and its subdomains. Browsing the site means accepting them. If you don't agree with them, don't use it.

2.2 Access

Access to the site is free and requires no registration. The documentation, the prices, the blog and the public demo are available without an account. An account is only needed to use the API and to subscribe to a plan.

Uttera is not responsible for the cost of your internet connection or for the equipment you use to access it.

2.3 Proper use

You agree to use the site in accordance with the law and with these terms. The following is expressly prohibited:

• Using it for unlawful purposes or in ways harmful to Uttera or to third parties.
• Introducing or spreading malicious code, or anything that could damage the systems.
• Unauthorized access, penetration attempts, vulnerability scanning, or mass automated extraction of content.
• Reproducing or transforming the site's content without authorization, except as allowed by the license of the published code.

2.4 Intellectual property

The texts, the design, the "Uttera" trademark and the logos are the property of Uttera or of its licensors.

The engine code is a different matter, and it's worth saying so here: the transcription and synthesis servers are published at github.com/uttera under the Apache 2.0 license, and you may use, modify and deploy them under that license, including commercially. The third-party models used with them carry their own licenses, stated in each repository.

2.5 Links

The site links to third-party pages (GitHub, Stripe and others). We are not responsible for their content or their policies, and a link implies neither endorsement nor a commercial relationship.

2.6 Availability

We make a reasonable effort to keep the site up and current, but the website itself carries no availability commitment. The availability commitment for the service — the SLA — depends on your plan and is in the documentation.

3. Terms of the service

3.1 What the service is

Uttera is an API that transcribes audio, generates speech from text, translates and summarizes recordings. It is subscribed to in plans with a monthly allowance of credits and billed per second of audio processed.

3.2 Your account and your key

Using the API requires an account and at least one key. The key is yours and it is your responsibility. Anything done with it is deemed done by you and billed to your account.

• Don't write it into source code, or into a configuration file that ends up in a repository, or into a phone system's dialplan.
• If you think it has been compromised, revoke it from your account page: the cut is immediate.
• You can restrict each key to specific IP addresses. It's the cheapest defense there is and almost nobody uses it.

3.3 Billing, renewal, cancellation and refunds

Payments are handled by Stripe Payments Europe, Ltd. Uttera never sees or stores your card details.

Automatic renewal. Your subscription renews automatically at the end of each period —monthly or annually, whichever you choose— at the then-current price, until you cancel.

Cancellation. You can change or cancel your plan at any time from your account. Cancellation takes effect at the end of the period you have already paid for: you keep your plan and credits until that date and then move to Free. There is no minimum term or penalty.

Subscription credits. The credit allowance is monthly and does not carry over: each month you get your plan's credits for that month, and whatever you don't use does not roll into the next month. On the annual plan you receive that monthly allowance every month for twelve months. If you exhaust the allowance, the service keeps serving you as billable overage, and the response tells you so with the X-Credits-Overage header.

Plan changes. An upgrade is charged pro-rata when you change; a downgrade is scheduled for the end of the current period.

Right of withdrawal (consumers). If you contract as a consumer —a natural person acting outside their trade or profession— you have 14 calendar days from the initial subscription to withdraw. By subscribing and starting to use the service you request its immediate performance. If you have not consumed any credits, we refund you in full; if you have consumed credits, the proportionate value already provided is deducted (email us at support@uttera.ai and we will settle it). Automatic renewal does not open a new withdrawal period: it counts once, at the initial subscription.

Refunds. Outside that period —or if you contract as a company or professional, who has no right of withdrawal— amounts for a period already started are not refunded; you keep your plan and credits until the end of the paid period.

Price changes. If we change the price, we will notify you before the next renewal and you can cancel before it applies.

3.4 Service level

The availability commitment depends on the plan and is published in the documentation. The free, startup and developer plans carry no SLA, and that is also stated on the pricing page before you subscribe.

3.5 Limitation of liability

Uttera provides an automated audio processing service. The results are automatic and may contain errors: a transcript can get things wrong, a summary can omit something relevant, and an acoustic estimate can be incorrect.

You are responsible for reviewing the output before making decisions based on it, especially where those decisions affect people. To the extent permitted by law, Uttera's liability is limited to the amount invoiced in the twelve months preceding the event giving rise to it, and does not extend to lost profits or indirect damages.

3.6 Suspension

We may suspend an account that breaches these terms, that puts the service at risk, or whose use is manifestly unlawful. Unless urgency or the law prevents it, we will give notice first and say why.

4. Prohibited uses

What follows isn't a defensive lawyer's list: it's what we don't want done with this. Using Uttera for the following is prohibited:

5. Cloned voice and recordings under Spanish law

This section exists because it's where most people get themselves into trouble without knowing it, and because we'd rather explain it than hide behind "the customer is responsible".

5.1 Whose voice you may clone

Yes, with peace of mind:

• Your own voice. To dub yourself into a language you don't speak, so you don't have to re-record a voiceover every time a sentence changes.
• Someone who has given you written permission, for a specific and bounded use: a hired voice actor with an express assignment, a colleague who agrees to voice the internal assistant.
• A voice that belongs to no real person, or one licensed for the purpose.

No:

• The voice of an actor, a presenter or a politician because public material exists. Accessible is not authorized.
• The voice of a customer or an employee because you have it recorded from a call. Recording for quality assurance is not collecting a voice model.
• The voice of a recognizable person, to make it look like they said something they didn't.

The practical test, which is almost always right: would the person that voice belongs to be comfortable seeing what you're about to generate with it? If the answer is "they'll never find out", you have your answer.

5.2 Which recordings you may process

Uttera processes the audio you send it; it does not check where it came from, and that check is yours.

• Recording a conversation you are part of is not a crime in Spain. Distributing it can be, and processing it for purposes other than those that justified it also has consequences.
• Recording a conversation between others, where you are not a party, is a crime (art. 197 of the Criminal Code).
• Recording calls in a company requires informing both parties in advance, having a legal basis and a defined retention period. If you also record your employees, see arts. 89 and 90 of the LOPDGDD.
• Special category data. A clinical recording is health data (art. 9 GDPR). Processing it is not prohibited — it is done every day — but it requires its own legal basis and, frequently, an impact assessment.

5.3 The duty to disclose generated audio

Since 2 August 2026, article 50 of Regulation (EU) 2024/1689, the AI Act, has been in application. As it affects you as a user of Uttera:

• If you generate audio that imitates a real person and could lead someone to believe it is authentic — what the Regulation calls a deep fake — you must disclose it clearly and visibly.
• If the synthetic audio informs the public on matters of public interest, the same applies.
• If your system converses with a person, that person must know they are talking to a machine, unless it is obvious.

That duty is yours, because you are the one publishing. Ours, as providers of the system that generates the audio, is to enable technical marking of synthetic content: article 50(2) requires outputs to be marked in a machine-readable format and the marking to be detectable.

5.4 Our marking timeline

These are the dates, so you can plan around them:

In July 2026 the European Commission published interpretive Guidelines — not binding — and there is a Code of Practice on Transparency of AI-Generated Content that providers may sign up to.

5.5 How we mark, since 16 September 2026

All audio Uttera generates carries a watermark, in both voices — standard and cloned — and in every format we serve. It is not optional and cannot be turned off: no API parameter allows it.

We use AudioSeal, by Meta, under the MIT licence. It is an imperceptible mark embedded in the sound samples themselves, not a file metadata field: it survives re-encoding, compression, trimming and format conversion. It is applied before encoding, so it travels in the audio you receive, whether wav, mp3, opus, flac or pcm.

Every audio response from our API also carries the X-Watermark header, declaring which scheme was used. If you ever receive audio from us without that header, that is a failure on our side and we want to hear about it.

5.6 How far the mark holds

A watermark is worth something if it survives the road the audio actually travels. These are our own measurements, on audio that came out of our engines:

We state it with its limit up front, including the case where it fails. A mark presented as infallible is not a mark, it is advertising.

5.7 On C2PA metadata

C2PA is a signed manifest that travels inside the file and that some platforms read to label content. We do not implement it, and we explain why, rather than leaving it as an open-ended "coming soon":

• It is not needed for compliance. Article 50(2) requires machine-readable, detectable marking. The watermark is exactly that, and it is in all five formats. C2PA is one way to comply, not the way.
• It protects against nothing. It is container metadata: any re-encode strips it, and so does any platform that transcodes. Anyone out to impersonate a voice is not going to send you a signed file.
• The standard does not support opus or raw pcm, two of the five formats we serve. Verified by us, not read off a documentation page.

C2PA does bring interoperability — a platform labelling the content by itself instead of treating it as unknown provenance — and that is real value. But it does not pay for itself: a manifest you can strip with one ffmpeg command proves nothing, and what has to be proven here is that a recording came out of a machine. The watermark does that, and it keeps doing it after a phone call. This is not a postponement: it is a decision. If the facts change — if the standard starts supporting opus, if platforms begin requiring it — we will revisit it and say so here.

5.8 How to check it yourself

We do not build a verification tool, and that is deliberate. We prefer something sounder: that you can verify it with Meta's public code, depending neither on us nor on us keeping anything running. A check is worth more when the interested party is not the one performing it.

pip install audioseal

from audioseal import AudioSeal
det = AudioSeal.load_detector("audioseal_detector_16bits")
prob, _ = det.detect_watermark(audio, sample_rate)   # mono audio, float32

That is the official AudioSeal detector, MIT licensed. Nothing of ours in between.