AI Vocalists Hit The Dancefloor: Who Owns The Voice

On 5 November 2026, KNTXT releases Azadi, a three-track EP from Charlotte de Witte. Every vocal on it belongs to NAVA, a singer who does not exist. NAVA was created in 2025 by Farbod Mehr, a London-based, Iranian-born artist, and according to Techno Airlines all proceeds from the EP go to FEMENA, an organisation supporting women human rights defenders. A black 12-inch follows around 11 December.

The project has a clear political purpose. In the EP statement quoted by Techno Airlines, de Witte says "NAVA represents Iranian women banned from public performance." It has still split her audience, and the argument around it is a useful way into a wider business question. As AI vocals move from novelty into records by some of the biggest names in techno, who owns a voice, who has to agree to its use, and how should it be labelled when the track reaches a DSP?

A Debate That Started In The Comments

The reaction to the teaser was quick. Magica, citing reporting by FAZEmag, noted commenters asking why de Witte had not given the opportunity to human singers, including Iranian women. Others defended the record as a way of bringing Persian vocals and poetry to an international audience. Mehr has said NAVA stands in for women who cannot sing in public in Iran, and that more than 70% of the views on one of NAVA's songs came from inside the country, although Magica points out that the report does not explain how this was measured.

Magica also notes that the published material does not say what NAVA's voice was built from. That gap matters more than the rest of the debate. For any record with a synthetic vocal, the first commercial question is not whether it sounds good. It is whose recordings trained the model, and whether those people agreed and get paid.

The Numbers Behind The Flood

Azadi arrives in a market already full of machine-made music. Deezer, which has published detection figures since January 2025, says AI tracks now make up more than half of the music delivered to it each day. According to TechCrunch, the streamer was receiving a monthly average of 90,000 AI tracks per day by June 2026, up from about 10,000 a day in January 2025.

Uploads and listening are two different things, though. In April 2026, when the daily figure stood at 75,000, Deezer said AI music accounted for only 1 to 3% of total streams, and that 85% of those streams were reportedly fraudulent, MusicTech reported. Deezer had tagged 13.4 million AI tracks since June 2025. "AI-generated music is now far from a marginal phenomenon," chief executive Alexis Lanternier said at the time.

In July the company said it would remove AI tracks that had not been streamed in six months, or that were tied to fraudulent streams. "Now that half of all daily uploads are AI-generated tracks, we are taking additional steps to safeguard the rights of artists and songwriters," Lanternier told TechCrunch.

The gap between those figures is the point. Most AI uploads are content-farm material that nobody chooses to hear. A record like Azadi is something else: a named artist, a real label, a release campaign and a vinyl pressing. The industry's rules were mostly written to deal with the first kind, and they are now being tested by the second.

What The Platforms Actually Allow

The rules depend on where the music goes, and they do not agree with each other.

Spotify: disclose it, and do not impersonate

Spotify's approach, set out in September 2025 and reported by Consequence, allows AI use but bans unauthorised vocal impersonation: a cloned voice is only allowed when the impersonated artist has authorised it. The company also backed AI credits based on the DDEX metadata standard, which let rights holders show where AI played a role across vocals, songwriting, production, instrumentation, mixing and mastering. "We support artists' freedom to use AI creatively while actively combating its misuse by content farms and bad actors," Spotify said.

Bandcamp: keep it human

Bandcamp went much further in January 2026. Its policy, covered by Synthtopia, states that "music and audio that is generated wholly or in substantial part by AI is not permitted on Bandcamp", and that "any use of AI tools to impersonate other artists or styles is strictly prohibited". The platform reserves the right to remove music on suspicion alone. Synthtopia read the policy as aimed at "AI slop" rather than AI tools used during production, but the wording leaves a grey area. A track whose entire vocal is synthetic could reasonably be seen as AI-generated "in substantial part".

For a dance label, that is a practical problem. The same EP might be welcomed with an AI credit on one platform and be at risk of removal on another, which is often where an underground release sells best.

Licensing Is Arriving, Slowly

The other half of the question is training data. In January 2026, Merlin, the digital licensing body for independent labels and distributors, signed a deal with Udio. Under it, Udio will build new models trained on licensed music from participating Merlin members. Merlin's announcement promises "fair compensation" and quotes Udio chief executive Andrew Sanchez saying artists will "maintain control over their work and are compensated for their creativity".

The announcement does not set out a rate, a payment structure or how individual artists are asked. That is typical of where AI licensing stands in late 2026. The framework exists, but the detail that decides whether a producer or a session vocalist actually gets paid is still mostly private.

What It Means For People Who Make And Play Music

None of this means AI vocals are off limits. It means a record that uses them now carries paperwork that a sampled acapella or a hired topliner never needed. A few habits are worth building now:

• Know the source. If a vocal comes from a model, find out what it was trained on and whether that use was licensed. If the tool cannot tell you, treat that as a risk.
• Get consent in writing. If a model is built on a real singer's voice, their agreement and their share of the income should be in a contract before release, not after.
• Fill in the AI credits. DDEX disclosure fields only work if labels and distributors use them. Check that your metadata says how AI was used.
• Check each store's rules. A release that is fine on one platform may break another's policy. Check before release day, especially for stores where your audience buys.
• Be open with your audience. The Azadi backlash was about choice and representation, not about hiding anything. Clear disclosure keeps the argument about the music.

For DJs, the change is quieter but real. Promos with synthetic vocals are already arriving, and the line between a cleared AI voice and an unlicensed clone of a famous singer will not always be audible. Playing a white label with a cloned vocal has always carried a risk. Now the clone is simply easier to make.

De Witte has chosen to put her name to the question rather than avoid it. Whatever people think of Azadi, the record makes plain that consent, credits and payment are now part of making a vocal track in electronic music, and the artists and labels that sort those out early will be in the best position.

Cover photo by dlxmedia.hu on Unsplash