Francisco Franco during speech at the swearing-in ceremony of loyalty before the Cortes loyalty to the Head of State of Prince Don Juan Carlos de Borbón on July 23, 1969.

“My dear infant, in the moments in which in compliance with the sixth article of the Succession Law, I take the decision to propose to the Cortes my successor in the Head of State in favor of your son Don Juan Carlos …”. Thus began the letter with which Francisco Franco communicated to Juan de Borbón the election of his son as future king of Spain. In the Xrey series, a podcast on the life of the monarch Juan Carlos I available now on Spotify, we heard it in the voice of the dictator himself. But Franco never read that letter aloud, at least there is no recording of that fact. It is an artificial intelligence algorithm that has generated the voice of the deceased to obtain a narrative resource – never before used in Spanish – in the podcast written, narrated and directed by Álvaro de Cózar and produced by Toni Garrido.

Deceased storytellers are nothing new. Twilight of the Gods by Billy Wilder began with the corpse of its narrator, young screenwriter Joe Gillis, floating in the pool of former star Norma Desmond. But now the resource goes beyond the narrative. Giving voice to historical figures has been done with actors and imitators, but today the technique allows us to go much further, it allows us to resurrect the voice of the protagonists of the story.

Technological capacity is not new either. In 2018, commemorating the centennial of his birth, the Times newspaper recreated the voice of John F. Kennedy to deliver the speech he was to deliver on the day of his murder on November 22, 1963. Created by the Rothco agency, the campaign JFK Unsilenced used artificial intelligence technology to recreate The Dallas Trade Mart’s speech: “We in this country, in this generation, are, by fate rather than choice, the vigilantes on the walls of world freedom.” Scottish company Cereproc took eight weeks to produce the 22-minute speech. They isolated each and every phoneme for American English in the voice of JFK and put these little speech units back together to get the words of the former American president. A complex puzzle only possible to solve with the help of artificial intelligence.

In the case of the voice of the Spanish dictator, the process has been very different. The Basque company Vicomtech, an applied research center on artificial intelligence and interaction in the Gipuzkoa Science and Technology Park in Donostia, developed a system capable of converting a text into a voice, not reusing phonemes, but learning from the speeches of the person whose voice wants to be replicated. This machine learning technology requires hours of recording to have enough material with which to feed the algorithm and that it can learn and train to accurately replicate the character’s vocal characteristics: tone, articulation, timbre and intensity, which can also change. throughout the person’s life, so the system should be able to reproduce that person’s voice at a precise moment in their existence. In the case of Franco’s voice, many registers were used.

The first of them, the speech of Salamanca in 1937, in the middle of the Franco civil war, the dictatorship that he plans to impose advances. The last one, only 20 days before his death, an umpteenth warning of the risk of Judeomasonic conspiracies. In between, hours of recordings recovered from the Internet and from private collectors who kept them on old vinyls. Many of them with poor sound quality and a lot of ambient noise, which is why the Vicomtech team explains that Christmas speeches are the raw material that has best served them to train the algorithm.

The result is a system capable of converting any text into the dictator’s voice. From letters of enormous historical importance to the lyrics of Macarena. A deep fake with enormous narrative possibilities but that also opens a debate on the use of technology in journalism and the media. What is the limit of doing interviews that never existed or listening to speeches that were never delivered? Where is the limit between history and fiction when technology makes them impossible to distinguish?