When AI finds a voice: keeping radio human in the age of AI
Radio has always been a medium built around voices. A familiar presenter can become part of someone's daily routine, a listener can recognise a personality within seconds, and a story can feel different simply because of the way it is told. That makes the arrival of AI particularly interesting for radio. When technology can generate, transform and reproduce voices, what happens to the human connection at the heart of the medium?
Two broadcasters explored this question from different angles. NGroup is experimenting with how AI can extend the personalities and relationships that make radio human, while Radio France is exploring how the same technology can transform and protect voices without losing their emotional qualities. Together, their work shows that the question is not simply whether AI can create a convincing voice. But what broadcasters choose to do with it.
What should remain human?
For NGroup, the starting point was not a technological question, but rather a human one. In 2024, the French-speaking Belgian radio group asked its teams to separate the parts of radio that AI could potentially take over from those that depended on qualities that are harder to reproduce: humour, spontaneity, empathy, local knowledge and the feeling of being accompanied by a real person.
That exercise changed the way the group approached AI. Instead of asking how much of a show could be automated, it started asking what technology could add to the relationship between radio and its audience.
One answer was to make a familiar personality available beyond the traditional broadcast. Brice 2.0 builds on Brice Depasse, a long-standing radio personality in French-speaking Belgium who has been sharing stories about music and artists on Nostalgie for more than 20 years. His archive was brought together in a structured database and connected to a conversational AI system, first through text and later through a cloned version of his voice. Listeners can therefore explore the knowledge he has built up over the years on air in a new, interactive format.
The approach can also begin with the listener. Le Labo Nostalgie turns listeners' own memories and experiences into a personalised listening experience. They answer questions about their lives and musical memories, and AI uses those answers together with the broadcaster's music database to create a first version of a personalised playlist. A human expert then refines the selection before it reaches the airwaves, making sure the music follows the emotional journey behind the listener's answers.
The response to Le Labo Nostalgie has been particularly striking. Listeners are often surprised by the playlists and, rather than simply naming an artist they want to hear, share personal memories that shape the selection. The format is still running every weekday at noon.
In both cases, the technology builds further on something that already exists: years of knowledge on one side, and personal experience on the other. But NGroup also explored what happens when AI is used to create something new.
Mamie l'IA was built from recordings of real grandmothers, combined with AI-generated knowledge and human performance. Several people were involved in bringing the character to life, from the original recordings to the writing and vocal performance. The result was a new radio character, rather than an automated replacement for a presenter. Because of latency and technical limitations, the program was also pre-produced instead of generated live.
The three experiments point towards the same shift: AI does not necessarily have to replace the voices audiences already know. It can make an existing personality more accessible, turn personal memories into a new listening experience or help create a character that could not have existed in the same form before.
But there is another possibility: sometimes the value lies precisely in changing the voice.
Changing the voice without losing the story
At Radio France, speech-to-speech technology is being explored for a very different editorial need: protecting people who need to remain anonymous. The system creates a different voice while retaining elements such as rhythm, intonation, emphasis and accent. This allows journalists to protect the identity of a source without stripping the testimony of the qualities that make it feel personal.
The contrast is revealing. In one setting, technology makes a personality or archive available in a new way. In the other, it deliberately removes the identifiable characteristics of a speaker. Yet both approaches depend on deciding what is worth preserving. For Ngroup, it’s the knowledge and personality attached to the speaker. For Radio France, it’s the emotion and character of the testimony.
That question becomes even more interesting when there is no original recording to preserve. Radio France has also worked on sound archaeology, using scientific data and AI to reconstruct what voices and sounds from hundreds of thousands of years ago might have sounded like, including those of Neanderthals, Denisovans and Homo naledi. Working with scientists, the team combined reconstructed voices with environmental sounds to create an audio documentary around a world that could never have been recorded. The project was later awarded Gold for “Best Use of AI”, recognising the way it combined AI with scientific research and sound design to bring the distant past to life
The team did not try to make these reconstructions perfectly human. Instead, some artificial characteristics were deliberately left audible, making it clear to listeners that they were hearing a reconstruction rather than an authentic recording.
There was also a much more personal application. Radio France recreated the voice of a sports journalist with ALS (amyotrophic lateral sclerosis) who could no longer speak, using one of his final interviews as the basis. Rather than inventing a personality or replacing a journalist, AI allowed an existing voice to be heard again in a new journalistic context.
What other media companies can take from this
- Start with the role you want AI to play.
Before introducing a new tool, identify what you want it to add to the audience experience or editorial workflow. The approach taken here shows the value of first considering what should remain human, and then looking at where AI can support it. - Build around what you already have.
AI can make existing assets useful in new ways. An archive, a recognisable personality, a format or audience contributions can all become starting points for new experiences. - Keep editorial judgement close to technology.
The examples show that automation does not remove the need for human decisions. Someone still needs to shape a personalised playlist, decide how a reconstructed voice should sound or assess whether synthetic speech is appropriate for a particular story. - Treat voice as an editorial choice.
Once voices can be generated or transformed, decisions about identity, emotion and transparency become part of the editorial process. The work at Radio France shows that altering a voice can serve a clear journalistic purpose, while also raising questions about what audiences should understand about what they are hearing.
The voice is changing. The editorial role is not.
What connects these examples is not that they use the same technology or serve the same purpose. It is that the reason for using AI comes from something human first: a relationship, a memory, a testimony, a personality or a story.
That makes the editorial question increasingly important. As AI becomes better at generating and transforming voices, broadcasters have more choices about what a voice should sound like, whose identity it should carry and how audiences should understand what they hear.
For NGroup and Radio France, AI is therefore not simply about making radio more efficient. It is opening up new ways of thinking about one of radio's oldest assets: the human voice.