Scarlett Johansson suspected Open AI of stealing her voice. In mid-May, the company introduced a new version of the chatbot, now capable of working with sound and literally speaking to the user. According to the actress, Open AI CEO Sam Altman had been persuading her for several months to voice ChatGPT, but she refused. After the presentation, Johansson claimed that the voice of the digital assistant “creepily resembles” her own.
In 2013, the actress voiced the character Samantha in the movie “Her” – a chatbot. The CEO of Open AI called this film his favorite, and a few days before the ChatGPT presentation, he wrote the word “Her” on his page. Here is how Johansson’s character sounds in the movie.
Scarlett Johansson’s lawyers sent a letter to Open AI demanding details on how the speech of the voice assistant was created. The version of borrowing from parts of the film looks very convincing, as modern neural networks require minimal original speech material to recreate a voice, noted Alexey Drozd, head of the company SearchInform’s information security department:
“From a technical point of view, it has long been shown that successful voice cloning can be achieved with just 3-5 seconds of original speech. Originally, about 300 phrases processed in a special way and in good quality were required to start. Additionally, they needed to be transcribed.”
Moreover, now synthetic voices can be created. Some companies offer to develop a branded voice assistant specifically for you, in which case there would be no need to deal with rights, royalties, and the like.
The mood, speed, and tonality of the voice can be changed, making it sound very natural as both male and female.
Open AI stated that the chatbot’s voice was not a synthesis of the American actress’s voice. The company representatives claim that a professional artist voiced the system, but did not specify who. Nevertheless, the voice from ChatGPT was removed out of respect for Scarlett Johansson. The American star has every chance to prove violation of her rights, according to Gleb Sitnikov, head of the intellectual property and personal data protection practice at the law firm “Yurlov and Partners”:
“It is necessary to understand that all this takes place in the jurisdiction of the USA. They have fairly flexible regulations in this regard, easily applying analogies of existing decisions to new situations. Even within the framework of Russian jurisdiction, I would probably try to argue that a person’s voice should be protected, for example, by analogy with the norm on image rights. This is a component of identity. Additionally, it can be considered personal data if explicit consent is not obtained. The voice can be biometric and can even demonstrate some harm, including reputational harm caused in this way. It may involve conducting an expertise, establishing the voice’s identity. If the dispute is not about this, but about them claiming that it is not a protectable object and can be freely used, then there would be no need to prove that it is her voice.”
In 2023, the largest strike by the Guild of Screenwriters in the US in 60 years occurred in part due to the use of neural networks in the film industry. Protesters demanded to stop using generated images of actors without their consent: deepfakes do not bring royalties, and background workers can be easily replaced.