HYBE is liquidating Supertone, its AI voice company. Who keeps the artists' voices?

Supertone, the AI voice company HYBE (the company behind BTS) used to make a singer sound fluent in six languages, is being wound up. Its shareholders voted to dissolve it at an extraordinary meeting on 15 July 2026 and appointed a liquidator, Music Business Worldwide reports. The decision was disclosed in a regulatory filing on 23 July. The filing page itself only records that a cause for dissolution has occurred, so the figures below come from MBW's reporting.

HYBE described the move as part of a realignment of its AI business strategy. Its sustainability report, filed on 30 June, still presented Supertone as a future growth business, according to MBW. About two weeks later the shareholders voted to dissolve it.

How much did HYBE put in, and what did it get back?

According to MBW, HYBE paid KRW 4 billion for an 18.2% stake in 2021. In 2022 it agreed to pay KRW 45 billion for majority control, and that deal closed in 2023. In total it invested nearly KRW 49 billion, about $35 million. In 2025 Supertone had revenue of KRW 2.3 billion (about $1.6 million) and an operating loss of KRW 15.4 billion (about $10.5 million).

The creative bets didn't land commercially either. MBW notes that neither MIDNATT's multilingual project nor the virtual group SYNDI8, which debuted in 2024 and drew criticism from some fans, became an established success. The most hopeful use was Big Ocean, a K-pop group of deaf members, who used Supertone's voice conversion on a single premiered at the UN's AI for Good summit in July 2025.

What happens to Supertone's products?

The company is being split. Production Expert reported Supertone's notice to customers:

• Shut down: Supertone Play, the API and Voice Builder stopped taking new sign-ups and payments on 23 July and closed on 31 August. Users got a bulk project export, and Supertone said remaining credits would be refunded automatically, in stages from early September.
• Transferred: Shift (real-time voice changing), Clear (voice separation and noise removal) and Air (dialogue matching) moved to Antinode Audio, a company based in Daejeon, South Korea. Supertone's notice told owners that their purchased licences and voices "remain unchanged".
• Personal data: customer data for those three products moved to Antinode on 14 September. Users who objected could ask to be excluded through a form before that date. Their accounts and data would then be deleted, and they might lose access to the products.

When Production Expert reported the transfer in August, Antinode's website was a one-page "coming soon" site. By 23 October antinodeaudio.com was offering Antinode Clear, Air and Shift and saying that existing licences and purchased voices remain valid. Supertone's own web addresses now redirect there.

Why does this matter beyond HYBE?

Supertone wasn't a minor plug-in maker. It was the label-owned AI voice business behind one of the most visible attempts to make AI vocals part of mainstream K-pop. On MIDNATT's 2023 single "Masquerade", HYBE used Supertone's tools to apply native speakers' pronunciation data to Lee Hyun's vocals in Korean, English, Japanese, Chinese, Spanish and Vietnamese, MBW reported at the time. The Korea Times reported that Supertone's voice design technology also built a female voice for the song from Lee Hyun's own voice. Big Hit Music's CEO said the technology could ease the burden on K-pop artists of learning foreign languages.

So the company being wound up has worked directly with at least two acts' voices: Lee Hyun's and Big Ocean's. Nothing we could find answers the obvious question. None of the filing, MBW's report, the customer notice as reported by Production Expert or Antinode's site says what happens to models or training data built from those artists. The customer notice deals with paying users' licences and personal data. It doesn't mention the artists whose voices the technology was built from.

The evidence from artists and fans is thin. Two X posts from K-pop watchers criticising HYBE's spending on Supertone were flagged to us, but we couldn't open them, so we haven't relied on them. We found no public statement from Lee Hyun, Big Ocean or their labels about their voice data. We'd rather say so than fill the gap.

What we think

The useful lesson here isn't "AI voice is a bubble". It's that an artist's voice model outlives the company that made it, and nobody at the shutdown is there to represent the artist.

Look at how the wind-down went. The tools that earned money (Shift, Clear, Air) found a buyer, with continuity promised to paying customers. The tools that closed (Play, Voice Builder, the API) came with exports and refunds. Users with a privacy objection got a deadline and a form. Each of these groups had a channel: a licence, a refund or a data-protection right. A singer whose voice was used to train or design a model is not a customer. Unless their contract says otherwise, they have no obvious seat in that process. And a liquidation is built to settle a company's assets and debts, not to ask artists what they want. This is our reading of how these processes work. We haven't seen Supertone's contracts.

There's a second-order effect for labels too. Owning Supertone kept HYBE's AI voice work inside the group. That control lasts only as long as the subsidiary does. Once a voice company is wound up, its models and data are assets like any other. They can be sold, kept, or quietly deleted, and the public may never learn which.

What we'd want any artist, or their manager, to ask for before their voice goes into a model:

• Deletion or return on shutdown, sale or change of control, covering the model, its derivatives and the training recordings, with written confirmation.
• Notice before any transfer, plus the right to refuse it. The opt-out Supertone gave its users should be the minimum for the people whose voices were the raw material.
• No assignment without consent. A buyer of the tools shouldn't automatically get artist-derived models along with them.

Supertone is a mild case. A large parent wound it down openly, with refunds, exports and a named buyer. The next AI voice startup to fold may not be that tidy. If these clauses aren't in contracts by then, the answer to "who owns my voice now?" will come from a liquidator, not the artist.