The new version of Djay Pro can isolate and extract different tracks from a music using an algorithm.
Digital music processing is certainly one of the most fascinating applications of computing over the past twenty years. And it could continue with the introduction of artificial intelligence algorithms for musicians, producers or DJs.
Among the features certainly most sought by the latter, the ability to “dissect” a song to extract a particular track, vocal, rhythmic or instrumental arrives in a good place, in order to bring variety to a mix. To be able to transform a passage from a song into a capella or instrumental version, or to play only percussion, is a rather tempting promise.
There are already services for this, which allow you to buy songs in a “source” version with four separate tracks, but the offer is relatively limited and expensive. Other means are used, using the sound frequency of a particular instrument, including the voice (which is one instrument among others). If, for example, in a song the frequency of the main voice is 200 Hz, tools can be used to precisely isolate this frequency and make it disappear from the song. The results are often quite convincing but not very “clean” with attentive listening. On the other hand, this method will remove all other notes and harmonics corresponding to the same frequency, because it does not distinguish between instruments.
Before remixing you must be able to “demix”
This is where artificial intelligence comes in. Just as our brain is able to tell the difference between different instruments (without having the ability to hear them in a totally isolated way), an algorithm is now able to act in real time to separate different tracks from a song.
This is what the new version of the Djay Pro app, aptly named Djay Pro AI, offers. With its Neural Mix technology and using Apple’s Core ML framework and the A12 Bionic or superior chip, the app is able to identify the different audio tracks of a music in real time while playing only the vocals, drums or instrumental part, with all possible combinations between these three sources, and this by simply acting on three zippers. The result is amazing, and even if it is not yet perfect or granularized enough because we are talking about three sources while a song is often recorded in 48, 64 tracks or even more, it is more accurate than with the method of isolation of frequencies.
In fact, this feature is comparable to what some image processing apps now offer that use AI to hijack a photo or blur its background by identifying and then separating the background from the foreground. These apps are therefore able to “deconstruct” a work from a single “flattened” file in order to extract the various elements and process them separately. This opens up infinite possibilities but can also raise questions about the creation and unalterable nature of a work as conceived by its author.
Djay Pro AI Neural Mix is available on the App Store for a monthly subscription of 5 euros. There is a free trial version of the Pro version valid for a week, which is more than enough to test the real possibilities.