The algorithm has once again been upgraded. The current version is now 0.3. The upgrade makes heavy use of feedback from two actual composers who reviewed it. User feedback (in the form of like/dislike) has also been incorporated.
Existing pieces that were generated using the old versions are retained. I hope the new version will be able to generate even better and more pleasant pieces. Just go to home page and try.
The Android app has been updated. Bugs were addressed and the user experience in the interactive mode (i.e. when you shake your phone to influence the generated piece) has been improved after usability tests were performed. You can get the app from the Play store.
Payments are now not available - working on a better and free way to get stock music from computoser
The algorithm has been updated. A lot of changes to the rhytmic and melodic rules have been implemented in order to improve the sound.
The Android app has been updated. It now has two modes - classic (simply plays generated pieces) and the new interactive mode, where you can influence the music you get by waving your phone around in the air. The motion is mapped to musical concepts such as tempo and variation and you hear pieces that correspond to your movements. You can get the app from the Play store.
Computoser now accepts Bitcoin (only if you need tracks for commercial purposes, they are free otherwise). First, why Bitcoin?
- there are no fees, there is no business entity verification process - you just set it up and start accepting payments
- computer-generated music is entirely "digital" - it doesn't require any human input or physical materials and so is Bitcoin. It made sense for a digital currency to be the means of paying for generated music.
And why I decided to have a payment option in the first place? All the tracks are free to listen to and licensed under Creative Commons. A .midi source file is available for download, thus allowing anyone to improve on the computer-generated piece. However, as the algorithm was recently getting better and better (far from perfect, of course), I thought that generated tracks can be used as stock audio. And when you get the mp3 and midi files, you can use them for commercial purposes (royalty-free, of course).
The Bitcoin integration itself has some caveats, which I'm going to describe in a separate, more technical article. There are multiple payment providers to choose from
, and I selected Coinbase. It's not perfect, but it's mostly straighforward and I got it working quickly.
Computoser is released - it is an online service that generates music algorithmically. Feedback has been very positive and this certainly means I'll continue improving it. But what could that service actually become?
First, as I noted in my technical description, the idea is not new at all. Many people, including scientists, have attempted to that the same thing - make the computer generate music. And some of these attempts (you can see some links in the discussions on HN and my blog) actually generate nice music, sometimes better than that of my algorithm. But why is it that this hasn't got any traction? Why there isn't an industry and a business model around these things?
- it's mainly research. Researches don't and don't have to think of business applications of their findings. Researchers are interested in the essence of the music composition and rarely in serving it to a large amount of people.
- the algorithms are not that good yet. Something can't get popular if it generates dissonant, boring or non-varying music. Mine also can still be classified as boring, but at least it's not dissonant and tracks may differ from each other significantly. Most of what I've seen is just recombination of a set of samples, which produces a limited set of results, or attempts to employ math principles, which I think is not necessary (mainly because existing, human-composed music doesn't seem to exhibit such patterns, apart from some low-level details, e.g. note pitch frequencies). Mine is far from perfect, but I've tried to address many deficiencies.
- existing software is not marketable - even if someone with a business plan took the software, he can't make it popular - the UI is in many cases horrible and unusable, and it expects a lot of human input. Something which is not actually needed.
- although it has been around, it is not an extremely popular idea. Not because it's bad, but because it requires a person or a team to have a good grasp on both computing and music theory.
But let's assume that I succeed in making the algorithm good enough and it generates very good music every time (for that, I'd need to analyze the likes/dislikes that people select for each track, consult musicians, etc.). Can it replace human composers and studios and performers completely? No. But it can generate tracks at low cost and high volume. The best composers will still be better, and some will benefit from the computer-composed tracks - by extending and improving on them, for example. And performing, especially singing, will obviously still be in business. (For the record, there are research projects that synthesize singing voices).
Why is all this needed in the first place, apart from scientific interest? Technology has made the current music business model obsolete - everyone can get the same track at no cost. And while that is considered illegal, enforcing the copyright laws goes through government control over the internet. And that's bad. The discussion about internet piracy is too big to fit here, but overall, technology advances, and so should the music business model
I mentioned business multiple times, but how can computoser and the likes be profitable? I'm not sure yet, but it is certain that end-users won't have to pay anything - otherwise it hits the same problem with the existing business model. Who is going to pay, can it be ad-supported, services like computoser will have to figure out. For now, all tracks generated by my algorithm are licensed with Creative Commons, the variation which disallows commercial use (it's not yet good enough for commercial use anyway). The difference is that there is no commission for composers, musicians and distributors - it scales because the web makes it scalable.
What if something like that succeeds? Can it transform the music industry? I won't jump into such conclusions so early. It may as well fail because there is no demand for that. But the idea itself has the potential to change the market. Let's see where it goes - in the future, or in the trashcan.