On 11 February 2026, nearly 1,000 people at the ElevenLabs Summit in London heard Patrick Darling sing a new song about his great-grandfather, a man he never met. Darling, then 32, could not sing it himself. He was diagnosed with amyotrophic lateral sclerosis (ALS), also called motor neurone disease (MND), at 29, and today he speaks very few words, with difficulty. The voice coming from the speakers was an AI copy of his own, made with ElevenLabs. Two of his former bandmates played live alongside it.

A musician losing his instruments

Darling has written and played music since he was about 14: bass and acoustic guitar, piano, melodica, mandolin and tenor banjo. But he told MIT Technology Review that singing was his "biggest love". In 2014 he joined the Ceili House Band, an Irish folk group led by Nick Cocking, as a singer and guitarist.

ALS took his abilities one by one. According to MIT Technology Review, it reached his legs first, and by August 2023 he had to sit to perform. Then he lost the use of his hands and could no longer play guitar or banjo. By April 2024 he was struggling to talk and breathe at the same time, and the band carried him onto the stage. Their last show together was in June 2024. In ElevenLabs' account of the project, Darling says losing his singing made him feel like a lesser version of himself.

Why voice banking was not enough

The usual advice for people facing voice loss is voice banking: recording speech while you still can, so that a text-to-speech system can use it later. A speech therapist suggested it to Darling. But by then his voice had already changed, and he felt he would be "saving the wrong voice", MIT Technology Review reports.

Richard Cave, a speech and language therapist and researcher at University College London, introduced him to ElevenLabs' voice cloning instead. Cave is also a consultant for ElevenLabs; the company's blog describes him as a member of its Partnerships team. Working from older recordings, they first rebuilt Darling's speaking voice, which he now uses to communicate every day.

A singing voice from noisy clips

Singing was harder. ElevenLabs' tool typically needs about 10 minutes of clear audio, and Darling had nothing clean: phone videos shot in noisy pubs and a couple of recordings of himself singing in his kitchen. According to a case study by AudioShake, a company that splits recordings into separate tracks, Cave used its software to pull Darling's vocals out of those mixed recordings, and ElevenLabs built the singing voice from the result.

The clone keeps the flaws of its source. Darling's singing was raspy and a little off on some notes, and the AI voice is too. Cave told MIT Technology Review that it does not sound perfect, but it does sound human. Darling describes it as sounding like him, but a different version of himself.

For the music, Darling and Cave spent about six weeks refining the track with Eleven Music, ElevenLabs' generator that creates songs from text prompts. Darling wrote the lyrics himself. The song is called "Ghost of a Man I Never Met".

Back on stage

At the summit, Darling stayed on stage while the track played, and his former bandmates Nick Cocking and Hari Ma played mandolin and fiddle live. His family were in the audience; according to ElevenLabs, it was the first time they had heard him sing in public since his diagnosis. Cocking told MIT Technology Review that the first time he heard the track he had to stop it, and it took him several attempts to get through it. After the concert Darling said he felt part of the band again.

In June, ElevenLabs reported that Darling performed another song, "Polish Girl with the Irish Heart", with the same singing voice at a SXSW London workshop held at Abbey Road Studios, and gave a live interview using his voice clone.

The caveats

This is a story ElevenLabs has every reason to tell. The first performance was at the company's own summit, the therapist who built the voice consults for it, and many details come from ElevenLabs and AudioShake, which sell the technology involved. MIT Technology Review, which interviewed Darling and his bandmates, is the main independent account.

The technology also has clear limits. Gabi Leibowitz, the speech therapist who leads ElevenLabs' Impact Program, told MIT Technology Review that it does not make communication faster and does not ease the physical symptoms of MND, such as problems with eating and breathing. What it can do, she said, is help people keep creating and working for longer.

How people with voice loss can get access

ElevenLabs runs an Impact Program for people with permanent voice loss. According to its healthcare page, approved individuals get a five-year, renewable ElevenLabs Pro licence for free and can create a clone of their voice; trained volunteers can help clean up recordings at no cost. The examples of eligible conditions it gives are MND, mouth cancer and multiple sclerosis. To apply, you create a free ElevenLabs account, open your profile and choose "Apply for Impact Program". Applications are reviewed by partner organisations, which include Bridging Voice, the Motor Neurone Disease Association, the National Federation of the Blind and the Scott-Morgan Foundation.

Bridging Voice, which helps people with ALS communicate, is the organisation that introduced Cave to AudioShake. Its website lists voice banking among its services, reports 3,032 voice clones created, and says it supports people in all 50 US states.

What this means if you or someone close needs a synthetic voice

Save the voice early. Darling's clone was built even from noisy phone clips, but the clearer the source, the better the result. If a diagnosis makes voice loss likely, record now: voice notes, videos, a few minutes of reading aloud.

Look for old recordings too. If the voice has already changed, old videos and voicemails can still be the raw material, as they were for Darling. Tools that separate vocals from background noise can make them usable.

Start with a speech therapist or a partner organisation. They can help you decide between voice banking and AI cloning, and connect you to free programmes before you pay for anything.

Pick the tool for the job. ElevenLabs is what Darling used for both speech and singing, and its Impact Program is free for eligible people. If what you mainly need is text read aloud in a natural voice rather than a copy of your own, a reader such as Speechify is built for that.