OpenAI TTS commercial use vs EasyVoice — Apache-2.0 license, AudioSeal watermark, consent gate

What license covers EasyVoice's Kokoro voices?

The Kokoro-82M text-to-speech model is licensed under the Apache License 2.0. Apache 2.0 is a permissive open-source license that explicitly permits: commercial use, modification (adapting the model weights), distribution (redistributing the model), private use (internal enterprise applications), and sublicensing. The license requires attribution (retaining the copyright notice and license text when distributing the model itself) and a patent license grant (contributors cannot sue you for patent infringement related to the model's use).

In practical terms: your rights to the audio come from EasyVoice's terms, not from the model license. You own the audio you generate; commercial use is allowed on every plan, including Free, with no attribution, royalty, or per-project license, and those rights survive cancellation (see /license). Apache 2.0 matters for a different reason — the Kokoro model itself carries no license restriction that could flow down to your product. The Arabic voices run on the Supertonic engine and cloned voices on OpenVoice; your rights to audio from those voices come from the same EasyVoice terms. What the terms do not allow without written permission is reselling API access or running a voice service for third parties built on EasyVoice.

Apache 2.0 license terms summary for Kokoro-82M: Permitted — commercial use, modification, distribution, private use, sublicensing. Required — attribution (when distributing the model), notice of changes. Prohibited — trademark use (using the Kokoro name as your product name), liability claims against contributors.

OpenAI TTS commercial use terms — what the policy actually says

OpenAI permits commercial use of TTS API output subject to its usage policies. Its text-to-speech guide (checked September 2026) states that the usage policies require you to provide a clear disclosure to end users that the TTS voice they are hearing is AI-generated and not a human voice. OpenAI's policies also prohibit using voice output to impersonate real, identifiable people without their consent.

OpenAI's policy does not restrict the commercial sectors you can use TTS in — advertising, entertainment, accessibility, IVR, and content monetization are all permitted. The restrictions are behavioral (what you do with the output) rather than categorical (what industry you're in). OpenAI also reserves the right to audit usage for compliance and can terminate API access for policy violations.

Comparison: EasyVoice has similar behavioral rules (no deceptive impersonation, no cloning a voice you have no right to), and its terms state that your rights to audio you have already generated survive cancellation. OpenAI's TTS API is a proprietary service whose commercial rights come from its terms of service. The Apache-2.0 license is a separate point: it covers the Kokoro model weights, not your audio.

AudioSeal watermark on cloned voices — what it means commercially

EasyVoice's voice cloning feature applies AudioSeal watermarking to all audio output generated using a cloned voice. AudioSeal is Meta AI's imperceptible audio watermarking system — the watermark is embedded in the waveform in a way that is inaudible to human listeners but detectable by AudioSeal's detection model. The watermark identifies the audio as AI-generated and can be verified programmatically.

For commercial use of cloned voices, AudioSeal watermarking means: the audio is commercially usable but carries a detectable AI provenance marker. If your product uses cloned voices for advertising, content creation, or brand voice applications, the watermark does not affect the user experience (it is imperceptible) but does establish a technical audit trail confirming the audio is AI-generated. This is a trust and accountability feature, not a commercial restriction.

OpenAI TTS does not publicly document watermarking on standard TTS output. OpenAI has developed C2PA provenance metadata tools for AI-generated content, but as of 2026, watermarking on audio.speech.create() output is not confirmed as a default behavior. If watermark detectability is a compliance requirement for your use case (journalism, content authenticity standards), EasyVoice's explicit AudioSeal watermark on cloned audio provides a documented, verifiable provenance trail.

Consent requirement for voice cloning

EasyVoice requires an explicit consent attestation before creating a cloned voice from an audio sample. The voice-clone endpoint rejects the request unless its consent parameter is set to 'true', and the attestation — your confirmation that you hold the rights to the voice — is recorded with the clone's enrollment. It is a recorded attestation backed by EasyVoice's terms and the AudioSeal watermark, not an identity check: EasyVoice does not verify who is speaking in the sample. Cloning comes with Pro or the 7-Day Pass, with up to 3 cloned voices per account.

The consent attestation requirement exists for two reasons: legal protection (non-consensual voice cloning of real individuals raises right-of-publicity and voice misappropriation claims in many jurisdictions) and ethical product design (voice is a biometric identifier and its cloning without consent raises identity harms). For commercial applications that want to build voice-cloning features, EasyVoice's consent gate is a compliance signal to regulators and a defensible product design choice.

OpenAI's standard TTS offering is its set of built-in voices. Its text-to-speech guide (checked September 2026) also describes approved custom voices for eligible customers, created from a speaker's consent recording and a matching audio sample — so on both platforms, a custom voice starts with the speaker's consent.

When does commercial-use licensing matter for your product?

Commercial use licensing matters most when: (1) you are distributing the audio output as a commercial product (audiobooks, branded content, advertising), (2) your legal or compliance team requires a clear open-source license on all AI model dependencies, or (3) you are operating in regulated industries (financial services, healthcare) where AI provenance and watermarking compliance is required. If you want to offer TTS to your own users — a platform, reseller, or white-label voice service — the self-serve plans do not cover it: EasyVoice's terms forbid reselling API access or offering a service to third parties built on EasyVoice without written permission, so contact us through the Send feedback button at easyvoice.ae for a written agreement.

Switch to EasyVoice if you want an Apache-2.0 licensed model behind your commercial audio, require AudioSeal watermarking on cloned voice output for AI provenance compliance, or need consent-attested voice cloning for your product's trust architecture. Stay on OpenAI if its built-in voices meet your needs and your legal review is comfortable with proprietary terms-of-service commercial rights (which most standard commercial products are) — OpenAI's commercial use terms are permissive for most standard applications.