A study, led by University of York researcher Mario Vallejo compared classic guitar amplifiers, software-based amplifiers and AI amplifiers. Surprisingly, they found that the AI amplifier was rated incredibly similarly to the classic one.
A guitar amplifier (amp) is an electronic device that boosts the weak electrical signal from an electric guitar, turning it into audible sound. It is therefore no surprise that an amp is the main player in an electric guitar’s sound. The Vox AC30, used on records by The Beatles, Queen and U2, is the most recorded guitar amplifier in the history of British music.
For years, software companies have tried to recreate amplifiers like the Vox AC30, so musicians can record without needing the physical hardware. But how good are these digital copies? If we can’t even get Eduroam to work, can a complex machine full of circuitry and vacuums be replicated by software?
The study, conducted by Vallejo and co-authors Michael McLoughlin and Gavin Kearney at the School of Physics, Engineering and Technology, compared a real Vox AC30 against three software versions.
Two were commercially available softwares, known to emulate Vox AC30s. The third was artificial intelligence that had been trained on recordings of a Vox AC30, in order to learn and reproduce its sound.
Twenty-two people took part in a listening test. They heard short musical clips of each and rated how close each one sounded to the real amplifier. The AI version was rated closest to the real amp more often than the others. Turns out human vs AI is not the real battle, its software vs AI!
The most groundbreaking result, however, is that in one of the three clips tested, listeners rated the AI version no differently from the real amplifier.
‘These tools can, in some cases, copy the sound of a specific real instrument so closely that the difference becomes very hard to hear,’ Vallejo says. ‘That gives musicians and producers access to high-quality sounds without needing expensive or hard-to-find hardware, which lowers the barrier for making music.’
This study provides a new angle to the debate that will likely define this century: can AI be the great leveler? Can software increase accessibility to those who are socioeconomically limited? As indie artists are on the rise, can more and more musicians create music without expensive hardware and even more expensive labels? Or is the Digital Divide the glass ceiling that will continue to leave marginalised artists out of the industry?
Vallejo also pushes back on the framing of AI as a threat to musicians, ‘Artificial intelligence in music is just another tool, not a replacement for creativity.’ AI-based amps will most likely be welcomed by the music industry for freeing up studio time. However, the threat of AI remains a question for the future. With increasing numbers of AI artists and bands being exposed online, we must ask the question: How much do we value human creativity?