The podcast series 'XRey' turns to Vicomtech to clone Franco's voice with Artificial Intelligence

This voice cloning constitutes a great scientific-technological challenge that Vicomtech has solved through artificial intelligence technologies applied to speech processing.

XRey, a series of The Story Lab in the form of an exclusive podcast for Spotify directed and scripted by Alvaro de Cózar, and produced by Tony Garrido, through ten 25-minute chapters, reviews the life of the King Emeritus Juan Carlos I.

Through unpublished audios, interviews and more than 40 direct testimonies, such as those of Rafael Spottorno, former head of the King's House; the Hispanist Paul Preston or the politician Alfredo Pérez Rubalcaba, among others, the most important moments in the monarch's life are revealed.

For the series, The Story Lab, he wanted clone the voice of Francisco Franco. Instead of going to an actor to imitate the dictator's voice, they put themselves in the hands of Vicomtech.

The Basque company has generated Franco's cloned voice exclusively for two key moments in the script and one of the two bonus tracks that complete the series, in which the creation process is explained.

The objective of this development has focused exclusively on responding through technology to a requirement of the script. XRey which consisted of reproducing Franco's voice in chapter 4, in which he reads a letter that he himself wrote to Don Juan proposing his son Juan Carlos as successor to the head of state, in addition to another intervention in chapter 5.

The technological challenge of this development based on Artificial intelligence has consisted of applying the cloning of a particular voice to an innovative narrative technique with a multitude of possibilities yet to be explored.

With the technology used and based on deep neural networks, initially located around twenty training audios to generate a quality model. However, in this case, the difficulty of finding audio in good condition, clean of noise and in the narrative style that was sought, meant that the model had to be generated with only 6 hours, composed mainly of the dictator's Christmas speeches.

With this limited material and the application of advanced Artificial Intelligence technology, Vicomtech has made it possible to generate a realistic speech synthesis model which finally acquires all the particularities, nuances and style of Franco's voice.

The technology developed has been the result of several weeks of work and the involvement of Vicomtech's Speech and Natural Language Technologies Research Group.

This week, Spotify has published a bonus track that explains in detail what the process of cloning Franco's voice was like.