There was a bit of controversy surrounding the decision to use a form of AI to revive John Lennon’s voice on what Paul McCartney called “the last Beatles record.” What they’ve done is a far cry from the crude AI imitations of artists littering Soundcloud today, and has more in common with a more mundane application of machine learning: noise reduction.
When you hear people talking about it, you’d think this is a ridiculous money grab that used modern speech synthesis technology to imitate one of the most famous songwriters of all time. But the real story is simpler and more moving than that, and the technology isn’t all that great.
As the band members recall in a tender short film about the song’s creation, ‘Now And Then’ was originally a piano demo made by Lennon shortly before his death in 1980. His widow, Yoko Ono, provided the tape. It was later recorded by the band, but the quality of the recording was not very good.
“When I heard ‘Now And Then,’ it was very difficult because John was kind of hidden in a way,” Starr says in the making-of featurette.
“Each time, I wanted a little more of John’s voice,” McCartney recalled. “This piano appeared and the picture became cloudy. Of course, at the time we didn’t have the technology to separate it.”
They “failed” trying to save the song in 1995, but in 2022 they were working on a documentary, Get Back, with Peter Jackson. The filmmaker and his team applied the latest audio processing techniques to the band’s archival footage to isolate individual instruments and voices.
“We paid a lot of attention to technical restoration, so that in the end we can take any soundtrack and split all the different components into separate tracks based on machine learning. It led to the development of technology,” Jackson says in the short story.
MAL, as they call it, is a version of audio isolation technology that has come a long way in the last few years. A machine learning model can, for example, be trained on many guitar tracks to learn what a guitar’s waveform or spectral signature is, and then extract it directly from the mixed track, with varying degrees of success.
It is now commonly used in video calls as well, using models trained on human voice. By suppressing everything except the speaker’s voice, you can eliminate background noises such as a barking dog or a noisy cafe in real time. A crude version of this was sometimes used to create karaoke versions of songs by identifying and removing the vocal track.
In the case of Lennon’s demo, it worked like a charm, as you can hear at the following URL: this timestamp Making of a short story.
“There it was – John’s voice, crystal clear,” Paul said. “Now we can mix it and make a proper record.”
Some may question the ethics of making that record, but John loved tinkering with technology, and of course he had originally written and performed this song with the intention of recording it, so everyone involved He seems to think that he would have been very supportive.
But more importantly, it seems to have served as a bit of closure for the group. The ups and downs of stardom and creativity they endured are more than well documented, but losing a friend and creative partner of decades in such a way puts this final loose end out of reach. It must have been torture hanging there.
As anyone who has lost someone can attest, every trace of that person becomes priceless. “Hearing John’s voice…that’s what we should cherish,” George Harrison said in 1995.
And now, with a quarter century’s worth of technological improvements coming to fruition, that’s exactly what they can do.
“That was the closest I’ve ever gotten to getting him back in the room,” Ringo said.
You can listen to “Now And Then” now here.