The Legality of Music Datamining

Question

Let us assume that a hit song or two from a list of, say, 200 artists, is transcribed into a digital format which allowed for melody/harmony and other types of analysis. These analyses would produce a set of statistical trends regarding melody shapes, interval jumps, and other information found from analyzing the aforementioned songs. This information would then be refined and used as the basis for a program that could help a user generate music for commercial use. Using only the suggestions provided by the program, it would be nearly impossible to completely recreate one of the songs used for analysis (unless the user intended to do so).

Where would this process stand legally? From what I've seen, academic research allows for some lee-way in terms of the content that can be studied. However, I haven't found anything that mentions whether or not one could freely use creative data (music you've bought and transcribed into a format which can be analyzed) to produce factual data (statistical trends) that could essentially be packaged into a product and sold. My gut feeling says that it wouldn't fall under fair use (even though technically the process should not detract from the sales of someone's music). Would you need a license for each song?

If this sits in a strange area legally, it makes me wonder how programs like Band in a Box were developed. Then again I suppose they could have paid session musicians to give them solos/melodies that they would then use for analysis in much the same way.

Any answers are appreciated!

Answer 1

The answer is not at all obvious. It seems that in the EU it is the default assumption that text data-mining requires consent for protected works, but there are specific statutory exceptions, and in the UK, there is an exception for research so that (p. 7)

researchers are allowed to make copies of whole copyright works that they have lawful access to, without asking for specific permission, as long as they are making these copies solely for the purpose of non- commercial research.

Therefore, I would conclude that under UK law, the proposed activity would not be legal (because it is not non-commercial, and also because it is not text). However, US law seems to be more open to data mining. The most applicable case would probably be Authors Guild v. Google, which found in favor of extensive "fair usage". However, Google Books are free, and the difference between "for commercial purposes" vs. "for non-commercial purposes" could well be the most heavily-weighted of the four factors. On the other hand, what you are proposing is vastly more transformative than putting searchable scans online, and the "effect on market" would arguably be positive (that is, no sales would be lost because a free copy was acquired instead -- the work itself would not be available).

However... we also treat music differently (there is a separate section of code, 17 USC 114). Bridgeport Music, Inc. v. Dimension Films, 410 F.3d 792 holds "Get a license or do not sample". Now it's true that the actual sound was being transformed and not analyzed into something else like a statistical profile, but still the amount copied was tiny and surely did not undercut sales.

It's not clear to me what this commercial product would do / is. If specific facts about specific works can be obtained, the possibility of a finding of infringement would be higher, compared to a programmer who discovers facts about existing music and then uses that factual knowledge to create a program that composers could use to create their own music. Facts are not protected by copyright; you may legally discover facts about music if you legally acquire a copy of the music. If your product simply embodies facts about musical expression (and is not just a re-coding of the actual expression), then copyright law would not be applicable.