Canadian Musician - May-June 2022 - Page 10








Former Google Executive Jessica Powell on AudioShake , using AI to deconstruct songs & the untapped potential for audio engagement

What if a mixed and mastered song could be uploaded and quickly deconstructed into high-quality guitar , vocal , bass , drums , and other stems ? It ’ s not a new idea , though until recently it wasn ’ t possible at a high enough quality to be commercially usable . But now , thanks to advances in artificial intelligence ( AI ), a number of creative and business possibilities are opening up . By combining technological know-how , a reverence for music creators , and business experience in Silicon Valley , AudioShake is leveraging AI to turn individual songs into many possibilities . But that is simply in the here and now , as CEO and Co-Founder Jessica Powell , a former Google executive , sees this audio-focused , AI-driven technology as the next frontier in social media and audio engagement .

Karaoke bars are not a common birthplace of music or audio innovation , but simple ideas can emerge from anywhere and , given time to gestate , turn into something quite interesting . But from a simple thought a decade ago of , “ Wouldn ’ t it be nice to karaoke to Gang of Four instead of ‘ Brown Eyed Girl ’ and ‘ Wonderwall ’ for the millionth time ?” has emerged AudioShake . This use of artificial intelligence to “ de-mix ” songs into stems is not wholly unique , though by many accounts , AudioShake is doing it the best . DJ Mag commended it for being the “ cleanest separation ” stem tech , and AudioShake won Sony ’ s 2021 Demixing challenge , besting competing creations by Facebook , Byte Dance ( TikTok ’ s parent company ), and others .
At the moment , AudioShake has a small engineering team led by its co-founders , Powell and her partner , Luke Miner . Powell began her career in the world of music rights working for CISAC , which represents SOCAN and the other authors ’ societies around the world , and eventually went on to the highprofile position of VP of communications for Google . She is also a critically-acclaimed author and a lifestyle , business , and tech columnist for the New York Times , Fast Company , Wired , and more . Miner is a music-loving data scientist and the co-founder of the music collaboration app Tunebend .
“ People have been trying to separate music for a very , very long time , and using AI and other technical approaches , particularly in the past 10 years , there ’ s been a lot of movement . We felt that with the state of artificial intelligence , we had gotten to a point where you actually could make a lot more inroads compared to what had been done in the past ,” explains Powell to Professional Sound . “ We spent about 18 months building
By Michael Raine
the technology to do that . The first few passes were terrible . I remember the very first song that Luke separated , it was an old Smiths song and it just sounded terrible . It was this demonic Morrissey voice and it was a mess — but it was intriguing .”
What the AudioShake team were after wasn ’ t something that was simply good enough for hobbyists and amateurs , but an AI-powered stem separation tool that was good enough for professional use . As mentioned , AudioShake isn ’ t the first or only company attempting this , with competition from Deezer-owned service Spleeter , LALAL . AI , and AudioNamix ’ s Xtrax Stems , among others . So , for AudioShake to succeed , the quality needs justify a business model that services professional artists , producers , and music editors , as well as labels , music publishers , and more .
“ We were very lucky , maybe about six months in , through a friend of a friend of a friend , we met Mary Megan Peer at Peer Music , which is a very large and independent music publishing company ,” says Powell . “ They were so generous with their time and they said , ‘ Look , we ’ ve been pitched this stuff before and it just isn ’ t commercial quality . It ’ s always amateur karaoke or some sort of hobbyist-type of thing , but nothing that we could use in a sync licence .’”
Kindly , though , Peer Music offered to provide songs to AudioShake to help improve its AI model and give feedback . “ They said , basically , ‘ Look , this isn ’ t totally there yet , but we ’ re interested in this . We think it ’ s the best we ’ ve seen so far . We ’ re going to send you some songs and why don ’ t you practice on them as you improve the model and we ’ ll give you feedback each time you practice ,’” says Powell , noting that “ Season of the Witch ” by Donovan was one of the Peer-supplied songs that