What does it mean when your own voice suddenly announces everything, from tram stops to porn advertisements?
Helena Hallberg is a singer and can also produce music herself. Microsoft was her first major client as a spokesperson.
Ryan Nava
The first time Helena Hallberg heard herself saying things she’d never said, she was scrolling through Instagram. There was a video from a musician friend that showed him working on a program that would give him the beat on stage through his headphones. A warm Swiss-German female voice announced the title of the song. Hallberg’s voice.
“I wrote to him straight away: ‘Do you know that that’s my voice?’ That’s how I became aware of all this,” Hallberg told the NZZ.
A few days ago she published a Tiktok video that touched a nerve. In it she says that she sold her voice for the AI assistant Cortana to Microsoft and thus lost control of it. “There is an online platform where you can let me say anything you want,” she says into the camera.
It was also funny when friends wrote to her saying they had just heard her in the burger joint or on the tram. “But it is also a reminder that I sold my vote for $3,000. And that sucks.”
Hallberg explains to the NZZ in more detail how the deal with Microsoft came about. It was 2020, during the pandemic. Hallberg is a musician. She has Swedish parents, but grew up primarily in Switzerland. With a scholarship, she moved to New York in 2016 to study at a music school. She met her husband, also a musician, and stayed.
They spent the 2020 lockdown in a two-room apartment in Manhattan. «All our gigs were canceled, the theaters were closed. . . The job came at the right time for me. I could really use $3,000.”
It is one of her first major jobs as a speaker. For the fee, Hallberg recorded around 1,600 sentences. To do this, she spent more than eight hours every day in her small recording booth for about two weeks. No background noise, no cracking, a very specific volume, exactly 10 seconds between sentences, those were the instructions. Every evening she sent in the material and then had to make corrections if a word needed to be emphasized a little differently.
Microsoft bans sexually explicit applications – actually
Hallberg not only sold sound recordings, but also the right to reassemble them as desired. It was clear to her that Microsoft wanted to use her as a virtual voice, for example as part of the Cortana assistant or as a voice for answering machines. Not that she would one day sell her voice on its platform as the Swiss-German female voice “Leni” to every interested customer.
There are advertisements on porn sites that are set to music in a Swiss dialect voice that sounds a lot like the “Leni” voice. That would be a violation of Microsoft’s rules. The company prohibits the use of the AI voice in “sexually explicit” applications. The tech company did not specifically answer the question of the extent to which this is controlled to the NZZ.
When asked how she deals with it, Hallberg waves it off: “I don’t know, I don’t even bother with it.” Of course, she also believes that Microsoft should at least make sure that something like this doesn’t happen. “But I can’t do much.”
In the USA it is easier to sell your vote
There are cases that give hope. For example that of speaker Bev Standing. She gave her voice to a Chinese translation service. Years later, she realized that her voice was being used on a new video app: Tiktok. Without any contractual basis. Standing sued Tiktok and agreed to a settlement with the company. Tiktok now has a different voice.
Things could be more difficult with Hallberg, after all, she signed a contract with Microsoft. The German music and copyright lawyer Georg Manthey believes their chances are rather worse if American law applies.
In the EU, something as personal as the rights to one’s own voice or face cannot be sold so easily, he says. Data protection rules and personal rights ensure that a contract must state very explicitly which rights will be sold, to whom and for what purposes. If the strict requirements are not met, there are various approaches to stop the use of personality traits.
In the USA, “buyout agreements” are much more common. Actors, for example, can use these to sell extensive rights to their bodies – for example for a video game. The legal situation in most European countries is less clear. This means more risk for companies and at the same time more protection for artists than in the USA. The chance of additional compensation is also more realistic in European legal systems. However, artists are not without protection under American law.
“It’s complicated, several countries are involved,” says Hallberg about her contract. She doesn’t want to reveal anything more. It can be assumed that Microsoft has secured itself well.
At least that’s what the company did to Irish spokesperson Remie Michelle Clarke. She also recorded her voice for Microsoft in 2020. She later discovered that her voice was being offered on the AI platform of another company that she had never worked with.
She told the British Telegraph that the lawyer who won the settlement with Tiktok also looked at her contract – and called it a “career suicide”. Microsoft can do whatever it wants with their voice.
It can still be lucrative to sell your voice and face
In a sense, Hallberg is lucky. Because her AI double is anything but perfect. “As soon as I read it, I reported back to Microsoft that some sentences in Swiss German don’t work because of the different grammar.” She still gets dubbing jobs from Swiss companies. The “Leni” voice is far from being competition.
This is also because there is little text and audio data in Swiss German that AI can learn from. Voices in English and Spanish are much easier to imitate. 30 seconds of recording is often enough for an imitation of acceptable quality.
The company Heygen produces such technology. This allows people to speak in other languages and say new things in videos.
Heygen is also behind a fake that recently caused a stir online. It looks like a normal influencer video. A young, pretty woman sits in the car, speaks enthusiastically into the camera and recommends wet wipes to combat the smell of sweat. The video appears natural and personal. Only very small errors disguise it as artificially generated.
This is AI.
It’s actually so over.
— Ate-a-Pi (@8teAPi) March 26, 2024
Even the woman featured in it initially believed the video was hers. She works as a professional influencer. It was only when she listened to the text that she realized it was an AI video, she told New York Magazine.
She deliberately gave her face and voice away to a marketing agency’s AI project, which made an AI version of her and sold it. Companies can buy realistic-looking videos in which they are praised exactly the way they want. She was surprised, but didn’t feel like she was being treated unfairly. At the moment she is even benefiting because the video has gone viral and made her better known. Many new customers now want to book the original.
The technology is not yet perfect. For important orders, companies rely on real photos of people. But that time could soon be over.
Most famous voices can earn a lot of money
Manuel Naranjo is a board member of the Association of Professional Speakers Switzerland. He gave his voice for a voice generator, an application for seriously ill people who can no longer speak for themselves.
The area of operation was clearly defined in his contract. “We recommend contracts in which every use of the voice must be renegotiated, and clauses so that the voice cannot be used differently even in the event of a company sale.” Naranjo received 7,000 francs for a week’s work, which is about normal for such jobs in Switzerland.
The most famous voices – like that of Arnold Schwarzenegger, who announces the directions for Garmin users – can receive hundreds of thousands of francs or more for one order.
Today, Hallberg also knows the general conditions, unlike four years ago, when she only began to build a foothold as a speaker in addition to her music. Finally, she speaks four languages fluently.
She is not a fundamental opponent of AI-animated voices. In the future, she hopes for platforms on which artists can sell their AI voice for music or lyrics, but are paid and retain control.
She is currently considering her options. After the video, she received a lot of messages with support on both an emotional and professional level, she says: “I feel less alone now.”