In just a few months, artificial intelligence has upset the calm in various creative trades. In the dubbing sector in Spain, the trigger jumped at the beginning of March. A studio in Madrid called on actors to “record emotions,” says Jaime Roca, president of the Voice Actors and Voice Talents Union of Madrid (AVTA), one of the nine industry associations in Spain. “We saw right away that it was to train an AI,” he says. “They make you record phrases with different emotions, which are then used for the machine to learn,” he adds. The computer is already capable of creating a new, artificial timbre, but it needs to copy the intonation: happy, sad, excited, enthusiastic, doubtful.
In the AVTA they have been, according to Roca, the first to release a statement. But they will not be the only ones. The nine Spanish groups have already created a forum on WhatsApp where they continue to talk and have already had meetings with Ibero-American and European groups in recent days. Soon there will be new collective communiqués, also international, with the aim of reaching Brussels. “We all have the same concerns and we want to create a lobby to inform politicians, it has been like a match that has fallen into a dry haystack”, summarizes Roca. The urgency and the desire to understand each other are proof of the magnitude of the challenge: “It’s a bit like aliens arriving,” she says. “Although he is never black or white, he has the ability to replace all of us and we will not be able to stop him. He will give opportunities, but it must be regulated, ”he demands.
In the statement, the union calls for, through regulation, the creation of something called an “AI accent.” Something that would make it possible to distinguish with a sound effect that a voice is not human: “Insert into the voice generation engine of all of them an equalization or sound effect to be determined that makes them identifiable at any time during their hearing,” claims the text. Roca clarifies that it would be like “a watermark, so you know that a program is talking to you, that it might sound like the radio,” he says. “This would avoid one of its dangers that goes beyond dubbing, impersonation and potential crimes,” he says.
The ‘human friendly’ label
“Something like this gives us the ability to compete, it would be like a label human friendly”, says Rock. These synthetic voices will soon have the ability to occupy any space. The dubbing of foreign actors, for example, could be done with the same timbre of the protagonist. The machine will be able to use Meryl Streep’s original timbre to make her voice be heard in Chinese, Arabic or Russian. The intonation in each language is what the voice actors now do, but the lips in the image would also be automatically adapted. If the AVTA proposal came to fruition, that voice would be heard with some sound effect that would identify it as artificial. “This would also suppress dubbing studies,” warns Roca.
There is a second essential point for Roca in a possible legislation: whose are the human voices that have served to train those artificial voices. This problem, which can end up in court, has also arisen with other professions, such as illustrators or writers. This massive collection of data is one of the reasons given by the Italian government to block the use of ChatGPT in the country: “We want to force the identification of all actions in its memory, which are not known, and charge for those rights”, says Rock.
The unexpected irruption of AI has put sectors that did not expect to be threatened in just a few months on the defensive: “I expected to see a robot waiter that would serve me the beers,” says Roca. “But making the machine is expensive, it’s cheaper not to make it and create a program that does other things, including creativity.”