Many voice actors at home and abroad publicly defended their rights: My voice was "stolen"!
"Day 44." On June 12, voice actor Lin Jing logged into Weibo and typed a line of words. More than a month ago, he clicked on a link sent by a friend and was surprised to find that the voice at the beginning of the AI sci-fi radio drama "Donuts from Another World" was 80% similar to his own. He suspected that someone might have "fed" the AI with his voice without authorization.
Coincidentally, not long ago, Zhejiang Sincere Law Firm was commissioned by a well-known voice actor who had dubbed for dramas such as "Temptation to Go Home" and "Princess Agents", and appointed lawyer Lin Yi Chong to send a lawyer's letter to the creators who plagiarized sound and other materials to produce and disseminate AI videos. The latter removed the relevant videos from major platforms.
Under the wave of AI, the dubbing industry is facing new challenges: the voices of some practitioners are "stolen" and fed to AI, which then takes away the jobs that originally belonged to them.
"A friend sent me this link in the early morning. I was so angry after listening to it that I couldn't sleep! I heard too many familiar voices, but according to the article, these voices are all classified as AI. The timbre and pronunciation of the first voice at the beginning are 80% similar to mine. Okay, even if mine is washed, what about the voice starting at 1 minute and 3 seconds? Is there any difference with Bloomberg's voice?"
Lin Jing and Peng Bo's Weibo rights protection journey began on April 30. They have dubbed game characters such as Yun Zhongjun in "Honor of Kings" and Zhongli in "Genshin Impact" and are also well-known in the industry. The two went to the notary office to record the screen to retain evidence such as the account and content of the post, and notarized it, making preparations for the final legal action.
Until today, Lin Jing has always posted a Weibo with the number 1, trying to use public opinion to put pressure on the other party. In recent days, the comment section has gradually gathered some voice actors who support AI and oppose the infringement of the rights of practitioners.
Voice actor Hong Haitian wrote: “Is it possible that any company or individual using human voices will have to clearly state the source of the voice data; each voice data must be labeled with the name of the worker and have the worker’s authorization?”
Faced with the menacing AI and ethical risks, similar scenarios are playing out abroad.
Recently, two American voice actors, Paul Lerman and Lina Sage, filed a lawsuit against AI startup Lovo, accusing the company of using AI to "copy" their voices without their permission.
Four years ago, Lerman provided a one-time voice sample on a gig website for $1,200, "for academic research purposes only." But recently, he heard his own "voice" narrating on a show. Sage's situation was exactly the same. She was paid $400 to make an advertising test script.
Scarlett Johansson, who plays the Marvel superhero Black Widow, was not spared. In mid-last month, when OpenAI once again attracted global attention due to GPT-4o, she issued a statement accusing ChatGPT of illegally using her voice and demanding that the voice be removed from the shelves.
Currently, OpenAI has suspended the use of this AI voice template.
"We, the online voice workers and video bloggers, are all high-risk groups." Xiao Mo couldn't help but sigh after seeing Lin Jing's Weibo. She switched from the media industry and has been working as an audiobook host in Hangzhou for nearly 10 years. She has seen some "traps" set for "voice theft".
“The market price for some jobs is around 300 yuan, and even if the sound is good, it is impossible to double it. But some companies suddenly offer 800 yuan, and there may be something fishy.” The “fishy” she was talking about is the "hidden clauses" in the contract. For example, in the rights and responsibilities section, hidden among a large number of copyright descriptions, there is a line of small print that mentions authorization for sound production and use, but the words such as AI do not appear directly.
The increase in such cases is also related to the fact that the technical threshold for AI to generate a person's voice is getting lower and lower.
Three years ago, Jianying began to launch text reading in a variety of voices, and at the beginning of this year, it launched the cloned voice function; Tomato Novel and Himalaya's AI audio novels are also very successful.
"These elements are actually glamrock, and then add this bling feeling. I think this dress has some jackets, such as the oversized jacket. I think I can be an off-duty model."
Recently, a text-to-speech tool called ChatTTS has become popular. As long as you input text, whether it is Chinese, English or a mixture of both languages, it can read it out in different tones, automatically add colloquial conjunctions, laugh, pause, and even imitate catchphrases.
"To customize a high-quality AI voice model, one or two hours of voice material and one or two days of training are required," said Zhang Zhengpeng, head of the long audio AIGC platform "Guagua Yousheng". Lingban Intelligence, the company behind it, has been focusing on the research and development and implementation of AI technology related to intelligent voice since 2016. It only takes a few minutes of material to complete an ordinary quality AI voice model. "The synthesis effect can be similar to the original speaker to a certain extent, but it is difficult to use in high-standard audio production scenarios."
A voice actor accidentally discovered that her voice had been converted into an AI and sold as a text-to-speech product on an app. She immediately filed a lawsuit in the Beijing Internet Court.
On April 23, the first instance court ruled that the defendant used the plaintiff's voice and developed the AI text-to-speech product involved in the case without obtaining legal authorization, which constituted infringement. The defendant was ordered to apologize in writing and compensate the plaintiff for various losses of 250,000 yuan.
As the first AI-generated voice personality rights infringement case in China, the verdict of this case has a groundbreaking demonstration effect in the context of the lack of relevant detailed guidance regulations. At that time, the court held that voice, as a kind of personality right, is personal and exclusive, and the voice of any natural person should be protected by law.
Article 1023, paragraph 2 of the Civil Code of the People's Republic of China stipulates that "the protection of a natural person's voice shall refer to the relevant provisions on the protection of the right to portrait." Zheng Yun, director of Zhejiang Sincere Law Firm, explained: "The legal provision is expressed as 'reference to' rather than direct application because voice is not a specific personality right and the rules of portrait protection cannot be fully applied."
"Legally, it mainly depends on whether the voice accused of infringement is identifiable. For example, if the public can identify that the voice accused of infringement belongs to a well-known voice actor, or it is enough to cause confusion among the relevant public, it may constitute infringement." Zheng Yun revealed that identifiability is closely related to the popularity of the voice right holder, and voiceprint comparison is one of the identification methods. "If there are more than 80% identical or similar, it may be determined by the judicial authorities to constitute substantial similarity and then be found to be infringement."
"If you feel that your voice may be infringed, you should collect evidence in a timely manner. Evidence collection methods include but are not limited to screen recording, screenshots, notarized screen recording and screenshots, etc." Lawyer Lin Yi Chong added, "Under the professional guidance of a lawyer, you can better complete the collection of evidence and all rights protection remedies."
In addition, in the opinion of the two lawyers, with the development of social and economic life and the increasing emphasis on personal rights, protecting the right to voice as a specific personal right will be the future development trend of our country's civil law.
While some voice actors feel uneasy or even angry because of their own experiences or because they hear or see the inappropriate use of AI, there are also people in the industry who are actively embracing the technology with their arms and trying to break boundaries.
On April 24, voice actor Zhao Qianjing authorized his voice to the AI audio drama "A Mortal's Journey to Immortality" for training multi-tone AI voices, which caused an uproar. Some people think that the involvement of AI is the general trend, while others think that this is a trampling on the art of dubbing and squeezing the living space of new voice actors.
At this stage, "mature voice actors and broadcasters are generally unwilling to make their voices into AI, especially big names." Xiao Mo said that those who are more willing to authorize AI sampling are usually newcomers with nice voices, who have little experience and skills but are recognizable.
Common practices in the industry include one-time purchase and split-use.
Not long ago, Xiao Mo authorized his son's voice to a company for a one-time buyout of 30,000 yuan. "It has been made into an AI voice, and the application scenario is similar to a reading machine. You can choose a sweet female voice, a majestic male voice, a male/female child's voice, etc." The little boy is about to enter the voice change period, and leaving this baby voice is also a kind of commemoration for him. "By the time he grows up, his voice will have changed a lot, and it will not affect his career in this field."
When it comes to AI technology, Xiao Mo admits that she feels a bit nervous, as if the wolf is coming. However, she and some of her colleagues know clearly that "we have no choice but to accept it. We can only keep learning and respond proactively." When she has free time, she will ask some companies about AI voice production and try to use it to do some auxiliary work.
Zhang Zhengpeng told her that the effects of AI voices have been improving, and many common broadcasting tasks are basically available, but in terms of complex emotional understanding and performance, there is still a big gap with real people.
Several personal experiences have also made her feel a little relieved: "Mature voice actors and anchors usually have unique styles, and they will adjust emotions and rhythms according to the different plots and writing styles." These subtle differences and emotional transmission cannot be replicated by AI at present. "High-quality works still need real people, which cannot be replaced. In terms of auxiliary recording, such as picture books and track matching, the AI error rate is still quite high at the moment, and manual inspection and adjustment are required." The entire industry is still in the stage of exploration and adjustment.
