Pop legend Karel Gott speaks from beyond the grave using AI
To celebrate Czech Radio’s centenary, its creative team wanted to highlight not only its vast historical archives, but also its future. And so the idea was born for Gott Forever, a project to recreate the speaking voice of Karel Gott, who passed away in 2019, using artificial intelligence. I spoke to Czech Radio’s creative producer Lukáš Sapík to find out more about the project, and started by asking: why an AI-generated voice?
“It was a surprise for the 100th anniversary of radio broadcasting in the Czech Republic. We wanted to prepare a special project focused on innovations and modern technologies, and voice synthesis was the ideal choice for radio, because sound is what we do and what we do best.”
And why Karel Gott?
“I think this project showed the importance of the radio archives. We have hundreds of hours of Karel Gott’s spoken voice in it, mainly from his own show, ‘I’ll Take This Movie Back’, which was broadcast by Czech Radio’s station Dvojka between 2011 and 2015. This was precisely the source we used for the AI-generated voice print, and there is no other famous person in the Czech Republic who has such a massive archive.”
How did his family react when you approached them with the idea?
“I think Ivana Gottová, Karel Gott’s wife, was surprised at first – she wanted to hear what the technology could do. The first samples were not yet 100% complete, but still she agreed that we could go ahead with the project and I think when she heard the final version, she was really moved.”
What about the ethical questions surrounding recreating the voice of somebody who is dead – how did you deal with those?
“We have some relatively strict internal rules, because there are no general rules made by the government or the European Union yet. For example, one of our rules was that we can use the AI-generated voice only for sentences that the person himself has written or spoken, and we must not change the context of those words. We also need to warn listeners that there will be an AI-generated voice in the broadcast.”
What can the AI-generated voice of Karel Gott do? I know it will be reading his autobiography, but can it also sing, for example?
[laughs] “No, it can’t. In fact, we haven’t really tried because the database of sounds that the AI used for the digital voice print is exclusively the spoken word. If we wanted to teach the AI to sing like Karel Gott, we would have to give it a huge number of singing tracks with no music, and we don’t have this material. So we have no plans for that right now.”
Was the database of sounds just him speaking Czech? I think he also spoke German – could the AI-generated voice speak other languages too? Even maybe languages he didn’t speak in real life?
“In fact, scientists all over the world are working on this right now to make it possible, and some models can already do this a little. Our model also has fairly good pronunciation, especially for English words, because there were many English titles and names in the programme ‘I’ll Take This Movie Back’, but our model is not ready for continuous conversation in a foreign language yet.”
Do you have any plans to create other AI-generated voices of famous Czechs? You said you have the most recordings of Karel Gott’s voice, but I imagine there must be some of Václav Havel, for example, or even Tomáš Garrigue Masaryk?
“The main problem is really the archive of sounds, because we need about 20 hours of recordings, and there are not many other famous Czech people which we can use for this because we don’t have such a massive archive for them.”
When and for how long can people hear Karel Gott’s AI-generated voice?
“It will be broadcast every day at 10pm on Czech Radio Dvojka for the whole of the holiday season, reading his autobiography ‘My Way to Happiness’.”