AI cloned my voice and stole my job.

Text | Sleepy、Menmen

Xiaoyu saw his own voice on a short video platform on an ordinary evening.

He had made a living with this voice for six years. The timbre, the pronunciation, the slight contraction of the larynx when taking a breath—it was exactly the same as his own voice.

The voice in the video said, "This life is really meaningless."

He stared at the screen, stunned for a moment. He actively avoided such depressing words when taking orders. He wouldn't say that, nor did he want to.

But that voice spoke with such self-righteousness. Xiaoyu turned voice acting into a full-time job at the end of 2020. His first job came from a relative; the relative's company had a science popularization short video project that needed voice acting, and knowing he had always liked it, they assigned the work to him. Later, he gradually entered the industry, mainly providing voice acting for short videos and self-media accounts, earning around ten thousand yuan a month. He caught the tail end of this business. Short video voice acting in China has grown into a huge assembly line, with the global voice acting market reaching over five billion US dollars in 2022. However, profits are shrunk further down the chain. Clients quote platforms about thirty yuan per minute, but by the time the voice actor gets their hands on it, they often only get one and a half yuan. Xiaoyu stands at the end of the chain. He has no breakout works, can't name many memorable roles, and can't be considered a proper "voice actor." He describes himself as someone who makes a living with his voice. His voice isn't a talent, nor is it artistic expression; it's his livelihood. He protects this livelihood very carefully. After taking on jobs for a while, he specifically reminds his collaborators not to share his contact information casually. He's afraid that overusing his voice will make it unrecognizable. An unknown voice actor doesn't have much to protect; scarcity is one of them. In November 2023, a friend sent him a message. Only then did he realize that someone online had cloned his voice to create AI voiceovers and was using it recklessly. He tried to trace where the voice came from, but to little avail. He'd taken on so many jobs in recent years that any video he posted online could be downloaded and his voice extracted cleanly using software. He didn't know which segment, by whom, or when. Things were getting worse. Partners quickly learned to use his AI voice to negotiate prices. Hiring a human voice actor, even at a very low price, would still cost tens of yuan. Deploying his own model, however, could reduce costs to almost free. This saved a lot of money for accounts that posted hundreds of videos a month. Xiaoyu even considered the other party's perspective, realizing that preventing them from using the app was, in a way, preventing them from making money. But when they were making money, they didn't consider that their actions were cutting off Xiaoyu's livelihood. More trouble came later. His cloned voice spread rapidly across the platform, and the review and algorithm began to directly classify it as AI. Content judged as AI couldn't enter the larger traffic pool. Once his AI voice became widespread, his real voice was also treated as AI and couldn't get in either. A real person is being dragged down by his own copy. His voice, as a form of labor, has been removed from him and is now encroaching on his own living space. Official media outlets have reposted videos made with his AI voice, and later even made a few themselves. The larger the organization using it, the more ordinary people believe that the voice is originally AI. When he seeks legal recourse, the larger the target, the more he seems like the liar. Even though he is the one being infringed upon, it looks like he's trying to profit from an "AI voice." The law isn't as omnipotent as he thinks. In April 2024, the Beijing Internet Court ruled on the first AI voice infringement case in China. Voice actor Yin's voice was used by a company for profit, and the court recognized this, stating that voice is part of one's personality, and AI-cloned voices require separate authorization before use, provided the user can still identify the voice. Yin won. However, Yin's victory stemmed from his ability to trace the source. His voice appeared on several apps, and the defendant was a specific company. Xiaoyu's voice, on the other hand, was scattered across hundreds of self-media accounts, with no traceable source or defendant. Each instance of misuse involved small-scale infringements of tens or hundreds of yuan. The law provides an answer for "one person suing one company," but not for "one person being misused across the entire internet." Xiaoyu coined a term for this relationship: parasite and host. Once a voice is no longer used in the traffic pool, users will seek out the next voice. "They'll switch to a new host," he said. "I wonder if the next host will last longer than me." The whip is lashing at another me. In the comments and private messages on Xiaoyu's personal account, people kept asking how to get his voice. "How do I use this voice?" "Your voice is really good, how do I clone it?" They asked with such confidence, as if they were asking how to download an app. This offense upset Xiaoyu. He didn't believe a voice could be so easily bought. In his view, if even a face and a voice could be completely bought, then a part of a person would become a slave again. When asked if the cloned voice truly represented "his voice," his answer was somewhat contradictory. He admitted that, in terms of similarity, it was his voice. He went to defend his rights and speak out precisely because he felt it was related to him. But he also said that it wasn't "him" because it wasn't with his permission or his consideration. He gave an example. "It's like someone is whipping another version of myself, forcing me to say certain things." We asked him if, if he really authorized it, the AI ​​voice wouldn't belong to him. He was silent for a moment. "Compared to not authorizing it," he said, "it belongs more to me." Being taken away, and being bought out. Neither is good; he could only choose the less bad one.

The path of licensing has already been explored to its end by some. In June of this year, Hasbro, the producer of "Peppa Pig," issued new contracts to the child voice actors involved, requiring the company to allow the use of their voices for AI replication and generation. After the media exposed this, agents of young actors in the UK issued an open letter of opposition, with only one reason: children simply cannot understand the long-term consequences of such licensing. More than a thousand people signed the letter. Making a child sign a license they can't understand is taking advantage of them, but even if an adult can understand the contract, what good is it? In the short term, it might bring in some money, but in the long run, they might not be able to control what the other party says or does with that voice. Moreover, when AI voice technology truly matures, the market will only favor the top performers. The voices of top voice actors will achieve a kind of immortality in the industry, making it impossible for younger voices to break in. Industry research shows that after widespread voice theft, over 70% of small and medium-sized voice actors lost more than half their income, and even top performers generally lost 30%. The remaining value of small and medium-sized practitioners is simply serving as training samples for models. "When that day comes," he said, "maybe the voice-over industry will be dead." The problem that plagues Xiaoyu is solving the problems of others. Helen is a documentary director and also makes commercials. She first used AI voice-over last year to edit a demo for a documentary project. To show the film to the client, she first needed to cut a five-minute short version, but the synchronous sound from the feature film couldn't form a complete short story. She had to rewrite a script and use voice-over to string the story together. She first found a voice-over company. With a limited budget, she chose a sample audio costing around 120 yuan per minute. Upon listening to it, the stability, rhythm, and emotion were all lacking, leading her to doubt whether it was even a real person dubbing it. Later, the editor suggested trying AI. After all, it was just an internal demo, not for broadcast or commercial use. The AI ​​voice turned out to be better than the cheap live voices from dubbing companies. This was the first time Helen clearly felt that AI dubbing had matured to the point of being usable. However, she was also clear about its limitations. It was usable for a demo, but not for formal commercial use, and certainly not for documentaries. The elements in the subject's synchronous sound—pauses, breaths, hesitations, the pauses before a sentence is uttered—are themselves part of the narrative. AI can process sound cleaner and smoother, but documentaries don't need smoothness. Imperfection is evidence of humanity. Helen paid for this imperfection. The algorithm also strips away this imperfection from Xiaoyu. The same thing can be creative material on one hand and a liability for traffic on the other. AI doesn't start by replacing the best voices. It first enters the gaps where "it doesn't have to be so sophisticated." Helen's demo is one such gap—opened reasonably, for internal communication, not broadcast, not for commercial use. Each gap is opened reasonably. But these gaps connect to form a road. On this road, the hardest to protect oneself is the person who finds it hardest to say no. Helen rejected AI, at the cost of having to make trade-offs between budget and timeline. Xiaoyu rejected it, at the cost of continuing to fight for her rights, continuing to waste time and money, and continuing to face the indifference and blocking of one account after another. The children who voiced Peppa Pig might lose a role altogether. And once a voice is cloned, it won't just be used for voice acting. In April 2023, someone in Fuzhou used AI face-swapping and voice-mimicking to impersonate a friend of an entrepreneur, swindling 4.3 million yuan in a ten-minute video call. In early 2024, an employee of a multinational company in Hong Kong was pulled into a multi-person video conference. The screen showed the headquarters CFO and several colleagues. Following the conference instructions, he transferred 200 million Hong Kong dollars. It was later discovered that, apart from himself, no one else in the entire meeting was a real person. What if one day, Xiaoyu's cloned voice were used to make such a phone call? What if his elderly parents received a call from "him" saying, "Life is so meaningless, lend me some money?" Now, every time Xiaoyu posts a piece of content, he has to do something that didn't exist before: prove that it is indeed him speaking. A real person, posting their own recording on their own account, has to first find a way to convince the reviewers and listeners that it is indeed them speaking. This process is clumsy, redundant, and shouldn't exist. This predicament was described by Wu Cheng'en more than four hundred years ago. In Chapter 58 of *Journey to the West*, the Six-Eared Macaque transforms into Sun Wukong. Not just "like" Sun Wukong, but every single hair is identical. The Demon-Revealing Mirror cannot detect it, and even Guanyin cannot distinguish it. The Bodhisattva Ksitigarbha's mount, Di Ting, listened intently from the ground and discerned the truth. But it dared not speak. It fears that if the fake voices cause a scene, no one can control them. This is so similar to platforms. It's not that they can't distinguish them, but that distinguishing them is pointless for them. Identifying a real person incurs verification costs and the risk of misjudgment. So, it's easier to just have AI for everything—the simplest solution. The most ingenious example is Tang Sanzang. The real Sun Wukong protected him throughout, but he had a bad temper, loved to talk back, and was prone to refusing to cooperate. The fake one was obedient, compliant, always available, and never threw a tantrum. Therefore, Tang Sanzang first drove away the real one. It's like those partners who use AI voices to pressure for lower prices. Real voice actors get tired, raise prices, and refuse certain submissions. AI voice actors never tire, providing whatever tone is requested, and keeping prices very low. From the perspective of demand, who wouldn't want to keep the more obedient one? Wu Cheng'en ultimately arranged for a Buddha to appear; only he in the Three Realms could distinguish between truth and falsehood, making a final decision with a single sentence—the false would be exposed, and the true would return to its rightful place. Xiaoyu's trouble lies in the fact that there is no Buddha in his world. Furthermore, a conspiracy theory has emerged among the common readers of *Journey to the West*, claiming that the blow at Leiyin Temple actually killed the real Wukong, because Wukong never threw a tantrum again afterward. This theory has circulated for many years because readers know that when only an authoritative voice can distinguish truth from falsehood, no one can guarantee that the blow was truly not a mistake. Even if Xiaoyu wins every legal battle, years later, the voice people remember may not be the one he uttered himself.