More than two decades after throat cancer silenced him, samba legend Cleber Augusto is singing once again. Warner Music Brazil has released Minhas Andanças, a new album that revives the voice of the former Fundo de Quintal member with the help of artificial intelligence.
The project combines AI-assisted restoration with human performance. Current-day singer Alexandre Marmita acted as a vocal stand-in, performing tracks that were then processed to sound like Augusto. The label hasn’t shared full technical details, but the approach mirrors recent efforts like the AI reconstruction of country singer Randy Travis’s voice in 2024.
The result is 13 re-recorded classics and a previously unreleased track, “Imã,” rescued from an old tape. It’s not just a nostalgia trip; it’s a quiet restoration of a voice thought to be lost forever.
“This project does not replace Cleber’s genius, but it allows his voice and his story to live on,” said Warner Music Brazil A&R manager Tony Vieira.
How to Restore a Lost Voice with AI
Tools like Elevenlabs, iZotope’s VocalSynth, Voicemod’s AI voice tools, LALAL.AI Voice Cloner or Respeecher can do the job, catering to different needs: some with real-time voice replacement, others with high-end studio focus. For example, Respeecher has been used in film and TV to recreate the voices of historical figures and actors.
LALAL.AI Voice Cloner lets users recreate a vocal model using existing recordings, such as studio sessions, live takes, or even old demos. Here’s how a process like Warner’s could be done outside a label system:
1. Gather voice data.
Collect as many clean vocal recordings as possible. These are the foundation for teaching the AI how the singer sounds details like vibrato, tone, accent, and breathing are key. LALAL.AI processes this material to train a custom voice model.
2. Create a donor performance.
Record a new performance of the track with a voice donor: someone who can deliver the right timing, energy, and phrasing. Think of this like casting an actor with the right presence.
3. Convert the voice.
Once the Voice Pack is ready, click on the three dots next to it and tap Use on Voice Changer. Then, upload the donor’s performance to LALAL.AI Voice Changer, pair it with the trained voice model, and let the software do the transfer. The final audio carries the donor’s delivery but sounds like the original artist.
4. Final polish.
The AI-generated vocal can then be mixed, mastered, and tweaked like any other studio take. Add effects or leave it dry, whatever suits the song.
These tools are now part of a growing toolkit for artists dealing with illness, aging, or just a lack of clean recordings. In some cases, they’ve also been used to bring deceased artists back into public conversation, though not without controversy. Cleber Augusto’s story is one of the quieter uses of this technology. But for fans, hearing his voice again isn’t just a technical achievement; but a return.
