If you have never had a musical bent, creating music may seem like some form of alchemy. A work of magic that only a select group of individuals with a very special set of skills, and a unique creativity, can ever master. So, to many, learning that a machine can create music - good music that to the untrained ear has nothing artificial about it - will no doubt seem like something out of a science fiction story. But that is exactly what tech startup called Jukedeck has done, as Breakthrough magazine found out.
In its simplest form, Jukedeck is an app that can generate pieces of music, based on a series of selectable styles and to any pre-defined length. Users are also able to select the point in the composition where the music reaches its peak, controlling the emphasis it can give.
The inspiration for the app came when Co-founder, Ed Newton-Rex, found himself accompanying his girlfriend to a computer science lecture - a subject he had had no previous exposure to.
A trained musician and published composer, Ed was blown away by the potential that computer science offered. Ed knew that music was essentially a very mathematical art and so an area likely to be suited to the very ordered world of computer science. He was also aware that Ada Lovelace - who is widely credited with writing the first computer programme in the mid-1800s - had prophesied that one day computers would write music.
Ed set himself the task of learning to code and to better understand the potential that algorithms offered alongside how they could be developed. Inspired by the potential to blend his passion for music with a new found interest in computer science, the idea for Jukedeck was formed and, in 2012, the journey began.
It would be three years before a working app would be released to the public - a period that saw a huge amount of development.
At the heart of Jukedeck are machine learning algorithms developed by Ed and his team. The algorithms analyse massive quantities of data from copyright free music of the style it is studying. From this analysis, it identifies statistical significances that help it learn how pieces in the music style are structured.
Analysing a series of classical pieces the algorithm may, for example, establish that 70% of pieces start with a tonic chord or that on 80% of the pieces, they end on a perfect cadence. From this, the system learns what was going through the composer's brain when the tracks were written, and it can develop rules that apply to the music style.
If it only studied a few pieces in each style, there would be a danger that all new tracks would be relatively similar in sound and style. But, the algorithm is trained with tens of thousands of pieces in each style.
The knowledge developed from this learning process is then used to shape the first stage of the creation of a new piece of music in the style - the composition. At this stage, the notes are created to make the piece of music. The sheet music if you like.
A significant part of the development journey has been spent identifying and preparing data for the app to use. The more data the algorithm has had to learn from, the greater its ability to create original pieces that suit the style.
Once the composition stage is complete, Jukedeck moves on to its second stage - synthesis. Here production systems developed by the Jukedeck team use other rules to add dimension to the original composition. This will include instrument choice and what sounds to use, what dynamics and accents should apply and other effects like reverb and compression.
The result is then converted to an mp3 file, ready for the app's user to download and use as they wish.
Computationally this is a hugely complex task that takes millions, if not billions, of calculations. And yet, a typical track will take around 30 seconds to generate.
But why did Ed want to create music in this way? An artificially created music track is never going to top the singles chart - is it? Well, it may well do. But that is not how the music is intended to be used.
If you have watched many videos online, particularly corporate ones, you will no doubt have winced at some of the backing tracks.
So many sound the same, and the quality is often far from good. That is largely down to the cost of creating or licensing unique, copy right free, music, for commercial use. Traditionally, good tracks have been expensive, beyond the budget of all but the largest companies. Smaller operations are forced to compromise on quality or choose something that was widely used, to keep the price down.
Of course, the use of video as a medium, both commercially and by individuals, is growing, fast - and that growth shows no signs of slowing down.
Jukedeck has grasped this opportunity in how it makes the app available. Organisations with less than ten employees can create and download a track for just 99 pence. For that, they get a unique piece of music that is of the exact length and style that their video requires. Something else which has traditionally made finding a suitable track difficult.
Larger organisations pay more, of course, but the JukeDeck team believe this is a fair way to approach the market. Those that can pay, and are likely to get more value from the piece of music, do.
The team has also opened up the technology though an API (application programming interface). Developers of other platforms can take the Jukedeck API and pull music into their own apps. This enables them to offer the benefits of Jukedeck to their customer base.
Overall, it's an approach that seems to be working well. Revenues in the business are split relatively evenly between the direct downloads and API agreements. While, on the download side of the business model, the proportion of tracks created by brands, as opposed to casual YouTubers, is showing healthy growth.
Since the app was formally launched at the end of 2015 over one million tracks have been created and downloaded. And Jukedeck has a client base that includes Google, Sotheby's and Coca-Cola. If it continues to grow as it has, and the commercial aspect of its customer base is expanded, then the business has a bright future.
So what’s next for the app?
The 20 strong team behind JukeDeck are currently working on a series of enhancements; functions that are designed to make the music even more valuable to users.
In the future, when creating a track, users will be able to control more areas including the selection of instruments, changing sections of a track and defining a narrative arc.
And will we ever see a Jukedeck track in the charts?
Well as suggested earlier, that may not be as far away as it seems. A YouTuber called Princess Rizu has taken JukeDeck tracks and recorded lyrics over the top. She has hundreds of thousands of views, and more YouTubers are following suit.
So, it's not a huge stretch to see it moving to the next level. It's likely to be when, rather than if. But when it does happen, do you think you will be able to tell the difference?
Judedeck has worked with Breakthrough funding, a company that helps UK SMEs achieve R&D tax credits - a government scheme created to enhance and reward innovation amongst UK businesses. Could you be eligible? Click here to learn more.
For more information, click here.