Warner Music Brazil Rebuilds Samba Legend’s Voice with AI. Here’s How Artists Could Do It Too with LALAL.AI

Two decades after losing his voice to throat cancer, Cleber Augusto is singing again. The legendary Brazilian samba artist has released Minhas Andanças, a new album featuring AI-assisted vocal restorations developed in collaboration with Warner Music Brazil.

Augusto, a former member of the celebrated group Fundo de Quintal, had not sung since undergoing surgery 21 years ago. But Minhas Andanças brings his voice back to life across 13 re-recorded samba classics and a newly unearthed track, “Imã,” written over 20 years ago and found on an old tape.

The label’s process, while not fully disclosed, reportedly involved a modern workaround: a current singer, Alexandre Marmita, served as a vocal “donor.” His performances were blended with archival recordings of Augusto’s original vocals to generate the new tracks. The result is a posthumous collaboration between past and present selves.

“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.

Warner’s experiment is similar to another recent project involving country singer Randy Travis, whose voice was reconstructed with AI in 2024 after a 2013 stroke robbed him of his ability to sing.

While Warner hasn’t revealed the specific tools used in Augusto’s case, projects like this are now within reach for many artists, thanks to tools such as LALAL.AI’s Voice Cloner.

Here’s how a similar process could be achieved using LALAL.AI:

How to Recreate a Lost Voice with LALAL.AI

LALAL.AI Voice Cloner allows people who suffer from throat cancer or ALS to restore their voices if they're about to lose them or already lost. In this tutorial below, we'll showcase how this works with the example of a singer's lost voice.

1. Voice collection & training

Start by gathering clean vocal recordings from the artist. These could be isolated vocals taken from past albums, demos, or rehearsal tapes. The more material, the better, ideally with varied pitch, emotion, and timbre. Then, upload them to LALAL.AI Voice Cloner to train a custom vocal model. This step essentially teaches the AI what the artist sounds like.

2. Voice cloning

Once the model is trained, it can replicate the artist’s voice with uncanny accuracy. The system will understand vocal nuances like vibrato, breath, and phrasing, not just pitch.

3. Voice transfer from donor

Next, record a stand-in performance from a voice donor, a singer who can deliver the track’s feel, phrasing, and musicality. Then, use the trained model to convert this performance into the voice of the original artist. The result is a vocal that carries the donor’s interpretation but sounds like the artist themselves.

For this, find the Voice Pack you've made on step 2 and upload it to LALAL.AI Voice Changer by clicking Use on Voice Changer.

Upload the song recorded with the voice donor's voice. Voice Changer will automatically replace its voice with the vocals from the Voice Pack. Once you're happy with the Previews, download the new song.

4. Mixing and mastering

The cloned vocal can be treated like any recorded voice: tuned, cleaned up, and mixed with instruments. Add effects, EQ, and dynamics processing to match the aesthetic of the original artist’s catalog or a new sound direction.

Whether reviving a lost voice or reimagining unreleased material, AI vocal tools like LALAL.AI offer a new way for artists, families, and labels to reconnect with voices that might otherwise be silent. And while AI won’t replace lived experience or the story behind a voice, it can help carry that story forward, one note at a time.

Follow LALAL.AI on Instagram, Facebook, Twitter, TikTok, Reddit, LinkedIn, and YouTube to keep up with all our updates and special offers.