Paul McCartney uses AI to rescue John Lennon's voice for final Beatles song

A noisy cassette demo abandoned nearly thirty years ago is finally getting a proper ending, thanks to artificial intelligence.

Sir Paul McCartney revealed on BBC Radio 4 that he has used AI technology to finish what he calls "the final Beatles record," scheduled for release later this year.

While McCartney did not name the track during the broadcast, it is widely understood to be Now And Then, a 1978 John Lennon composition that the surviving band members tried and failed to complete decades ago.

George called it rubbish in 1995

The song traces back to a lo-fi cassette labelled "For Paul," recorded by Lennon on a boombox while sitting at his piano in his New York apartment shortly before his death in 1980.

Lennon's widow, Yoko Ono, handed the tape to McCartney in 1994. Producer Jeff Lynne successfully cleaned up two other tracks, Free As A Bird and Real Love, which were released during the Beatles Anthology project in 1995 and 1996.

Now And Then, however, was quickly abandoned after just one afternoon of work. Lynne recalled that while the song had a chorus, it lacked completed verses, and the background hum from Lennon's apartment wiring made the vocal extremely difficult to salvage.

McCartney previously told Q Magazine that George Harrison flatly refused to continue, calling Lennon's vocal sound quality rubbish. Because the Beatles operated as a democracy, the project was shelved indefinitely.

How Peter Jackson unlocked the vocals

The breakthrough arrived during the production of Peter Jackson's Get Back documentary.

Dialogue editor Emile de la Rey trained a custom machine learning model to distinguish the Beatles' voices from background chatter and instruments. McCartney explained that the system allows sound engineers to instruct the software to identify the voice and separate it cleanly from the piano.

Jackson's team applied this stem-separation technique to Lennon's rough cassette, extracting an isolated vocal track pure enough to be mixed like a modern studio recording.

The same audio technology previously allowed McCartney to perform a virtual duet with Lennon on his recent tour, and powered the surround sound remix of Revolver released last year.

Still, McCartney admitted that broader AI developments make him uneasy, noting that hearing AI-generated tracks of Lennon singing other songs feels both scary and exciting.

Why it matters

This project represents the most tasteful and consequential use of AI in heritage music so far.

We think the crucial distinction here is that McCartney is not using generative voice cloning. Lennon is not being synthesized or faked; instead, AI is functioning as an ultra-precise digital scalpel, carving an authentic 1978 vocal performance out of tape hiss and mains buzz.

For the music industry, this opens up an entirely new revenue stream and preservation standard. Decades worth of historically vital home demos, bootlegs, and live rehearsal tapes that were previously unusable due to bleed or noise can now be restored to commercial release standards.

Yet the ethical questions will not go away. Harrison explicitly disliked the track and vetoed it when he was alive. While fans will treasure hearing the four Beatles play together one last time, polishing an artist's unfinished bedroom sketch post-mortem remains an inherently controversial creative choice.