The Beatles speak, machine learning listens

Despite their separation in May 1970, the Beatles are regularly in the news as Christmas approaches. This year, it is thanks to a film of almost eight hours by New Zealand filmmaker Peter Jackson, “The Beatles: Get Back” available in three parts from the end of November 2021 on the Disney + video platform.

The fruit of four years of work, complicated by the pandemic, this document is dedicated to a period well known to fans: in January 1969, in a climate of high tension, the group tried to produce an album in concert conditions, as in the beginning. They rehearse at London’s Twickenham Film Studios, while a film crew led by Michael Lindsay-Hogg films sessions on the spot. These will give rise to the documentary “Let it be”, released shortly after the separation of the Beatles. The record project, which was to be titled “Get Back”, is discontinued. It will never see the light of day as intended. It is in this context that the group will end up playing on the roof of the building of its record company, Apple Corp., on January 30, 1969.

Unintelligible conversations for 50 years

Peter Jackson collected all the footage shot by Michael Lindsay-Hogg over 21 days and restored it. But his true performance is having managed to make conversations between musicians audible, unintelligible for 50 years. To do this, he used machine learning algorithms.

In January 1969, in fact, the film crew used a nagra and moved around the rehearsal studio with a suspended microphone to capture sound on the fly, in cinematic style. Michael Lindsay-Hogg hoped to capture snippets of conversations between the Beatles, even hiding microphones near the musicians. From there, a game of cat and mouse was launched. Whenever the Beatles wanted to chat without being heard, they would turn on their amps and roar their guitars to prevent the microphones from having their voices.

Since all the sound ended up directly on a nagra tape, there was only one track and there was no mixing, everything was recorded in one piece. So Peter Jackson hired machine learning specialists at his post-production company Park Road Post. They have trained an algorithm to recognize the sound of a guitar. With the sound of the files converted to digital format, it becomes possible, working on a computer with this technology, to detect and isolate only the guitars to eliminate them from the soundtrack or lower their sound level, independently of the rest. In short, Peter Jackon’s teams artificially created a multitrack sound from a mono signal.

Unmixing, a technique developed since 2016

This “unmixing” process did not stop there. Once the technique is developed, it can be applied to any instrument: piano, bass, drums, vocals… This is exactly what Peter Jackson’s teams have done. All the instruments could be isolated each one in a file, each constituting, again, a track that, originally, did not exist. Thus, the music of the documentary could be restored but also remixed correctly, instrument by instrument.

“The Beatles: Get Back” does not mark the beginning of unmixing. The technique was developed from 2016 thanks to a computer technician from London’s Abbey Road studios. It had already been developed for an archive of The Beatles, the record in concert at the Hollywood Bowl. The process was then applied to David Bowie or the Rolling Stones recordings.