Skip to content
Euphona

Legal

Acceptable use policy

The short version

Process audio you have the right to process. Do not use the vocal tools to impersonate someone. Do not attack the platform or resell it as your own. Enforcement escalates — warning, then suspension of processing, then termination — and there is a real route to disagree with us.

A summary is not the agreement. Where the two differ the clauses apply — but if the summary is misleading, that is our mistake to fix rather than yours to discover.

What this covers

This policy applies to everything: Euphona Studio, the API, and the free tools on this site. It is part of the terms of service.

It is short on purpose. A policy nobody finishes reading is one nobody follows, and almost everything below follows from two ideas — process audio you are allowed to process, and do not use this to hurt people.

The rights warranty

When you submit audio — in the Studio, through the API, or into a tool — you are stating that you have the rights necessary to have it processed. That includes the recording, the composition, the performances on it, and any rights belonging to the people whose voices are in it.

If you are calling the API on behalf of your own users, this obligation is yours: you warrant it for the audio your integration sends, and you need the corresponding promise from them.

Processing a recording does not create rights in the output. A stem separated from someone else's record is still someone else's record.

What you may not do

  • Infringe copyright, particularly at scale. Processing a track you do not have the rights to is a breach. Doing it systematically — a pipeline that separates a commercial catalogue, a service resold on top of ours for that purpose — is the version that ends an account rather than warns it.
  • Impersonate a real person's voice to deceive, defraud, harass, or put words in their mouth. See below; this one deserves its own paragraph.
  • Process material obtained through unlawful surveillance — recordings made without a legal basis, intercepted communications, or covert recordings where the law required consent.
  • Produce material that sexualises children, or that constitutes unlawful harassment, incitement to violence, or content that is illegal where you are. This is the one category with no ladder attached: it is reported and terminated.
  • Break, probe or overload the platform. No attempting to extract our models or their parameters, no circumventing quotas or rate limits, no using outputs to train a competing system, no automated abuse of the free tools or the free tier.
  • Create multiple accounts to obtain more free minutes. The free allowance is once per person, not once per email address you control.
  • Resell the service as your own without an agreement. Building a product on the API is expected and welcome; presenting Euphona as your own processing engine is not.

Voices and likeness

Vocal processing raises a question copyright does not answer. A voice is identifiable in a way that a chord progression is not, and the harm from misusing one lands on a person rather than on a rights-holder.

So, explicitly: do not use these tools to make a real, identifiable person appear to say or sing something they did not. Not as a joke about a public figure, not as a demo, not as a proof of concept. Consent from the person whose voice it is makes it fine; the absence of a specific law where you live does not.

In the United States this is now law in a growing number of states, and not the kind that only regulators enforce. Tennessee’s ELVIS Act protects a person’s voice as a property right and lets them sue over an unauthorised digital replica. California protects living and deceased performers alike, with damages, and voids contract terms that purport to license a digital replica without informed, represented consent — so a release someone signed without understanding it may be worth nothing. Other states have followed and more are moving. A federal bill exists and has not passed; nothing about that helps you, because the state laws already bite.

There is EU regulation too. Since 2 August 2026 the AI Act has required whoever publishes AI-generated or manipulated audio constituting a deep fake to disclose it clearly, at the point someone first encounters it. That duty sits on you as the publisher, not on us, and it applies whether or not you had our permission.

Our prohibition is broader than any of them: disclosed or not, lawful where you live or not, do not put words in a real person’s mouth without their consent.

If someone has done this with your voice, or has separated your performance out of a recording you did not license, tell us at our copyright and voice page. You do not need to own any copyright in the audio to complain — the right in your voice is yours whoever owns the recording, and we handle those complaints on that basis.

Using the platform fairly

Rate limits, quotas and spend caps exist and are published. Working within them is expected; working around them is not. If you need more, ask — a raised limit is a conversation, and it is a shorter one than an enforcement action.

The free tools are free with no account and no email, which only stays possible while nobody automates them. So requests to them are counted per IP address and limited. For the browser-based tools that is the only thing we log about a tool you use, since the audio itself never leaves your browser; for the vocal remover the upload itself exists on our servers for up to twenty-four hours, and the count is kept in memory rather than as a record of who uploaded what. The privacy policy says the same thing from the other side, and it says it because an IP address is personal data and a page claiming the tools collect nothing at all would have been wrong. If you need programmatic access, that is what the API is for.

How this is enforced

The ladder is: warn, then suspend processing, then terminate.

  • A warning says what we saw and what has to change. Most cases end here, because most breaches are someone not having read this page.
  • Suspension of processing stops new jobs while leaving your account, your files and your ability to export them intact. Suspension is not deletion, and we do not hold your work hostage over a dispute.
  • Termination is last, and follows the account-closure process — you keep the right to export, except for the specific material subject to a valid claim.

We skip rungs only where the law requires it or where waiting would cause real harm. Every action taken under this policy is recorded — what was done, by whom, and why — in a log that cannot be edited afterwards. That record exists to constrain us as much as to document you.

Whenever we restrict anything, we tell you why in writing, and the notice contains the same things every time: what we saw and when; which term or which law it breached; whether a machine or a person decided it, and if a machine, that you can have a person look again; what exactly is restricted and for how long; and how to contest it. EU law requires that of a service like ours, and we would send it regardless — an enforcement notice that does not say what you did is indistinguishable from an arbitrary one, and it leaves you nothing to answer.

To report something illegal here that is not a copyright or voice matter — those have their own procedure — write to legal@euphonaai.com with enough detail to find the material and to understand why you say it is unlawful. You do not need to be affected by it, you do not need a lawyer, and you will get a reasoned answer rather than a ticket number.

If material processed here infringes your copyright, send a notice to copyright@euphonaai.com, including: what work is infringed, where the infringing material is, how to reach you, a statement that you believe in good faith the use is not authorised, and a statement made under penalty of perjury that you are the rights-holder or authorised to act for them.

What happens next depends on the surface, and the difference is real:

  • Stored material in the Studio — access to the specific assets is disabled, the account holder is notified, and they are told about the counter-notice route.
  • API jobs — the notice is forwarded to the key owner and their repeat-infringer clock starts.
  • Free tools — nothing persists beyond twenty-four hours, so there is usually nothing left to remove. We will say so plainly and document the notice rather than perform a deletion of bytes that are already gone. Pretending otherwise would be theatre. Where the uploader signed up and kept the result, it is a workspace file and the Studio entry in this list applies.

If you think we got it wrong

A notice is an allegation, not a finding. If material of yours was disabled and you believe the claim is mistaken or the use was authorised, send a counter-notice to the same address identifying the material, stating under penalty of perjury that you believe it was removed in error, and giving contact details.

We will pass it to the complainant and, unless they begin legal proceedings within the period the law provides, restore the material. This route is not a formality we hope nobody uses — bad-faith takedowns are common enough that a platform without a working counter-notice process is simply choosing a side in advance.

Repeat infringement

Repeat infringement is counted per workspace and ends in termination. We do not publish a precise number, because a published threshold is a budget — but the count is real, it is recorded, and warnings say where you are in it rather than leaving you to guess.

Questions about this policy, or about an action taken under it, go to legal@euphonaai.com. If you think an enforcement decision was wrong, say so — decisions get reviewed by a person, and some of them get reversed.