In the following column, PPL CIO Mark Douglas examines the contribution being made to the industry’s data challenges by Music Recognition Technology, its potential to be applied further – and the need for caution along the way…
Audio fingerprinting, or music recognition technology (MRT), is an established technology offering that, in recent years, has been developing into new services. Services that present some interesting opportunities for the industry but which I think need very careful consideration.
MRT clearly already delivers significant value to our industry and is a technology that we have been using for well over a decade at PPL.
With some of its newer potential uses, however, it may not quite be the silver bullet that many claim it to be. The challenge for rights administrators like PPL, as with crypto technologies, is to look beyond the marketing hype and ensure that any new technology we adopt delivers value to our members at a proportionate level of cost.
Global industry body, IFPI, maintains a tracker of all the companies that operate in this space, and the last time I enquired, there were around 60 companies on that list. You may well have heard of some of them: Gracenote, Audoo, DJ Monitor, Soundmouse, BMAT, ACRCloud, Audible Magic and Radiomonitor are just a few of the more prolific examples.
It is a service we use at PPL not just to source playlist data for our own licensees, but to use as an audit tool to check the completeness and accuracy of playlists we receive from a variety of sources.
Building on TV and radio beginnings, and in response to the huge growth in the use of music in social media and video streaming services, MRT’s next port of call was the internet.
Scanning the content of these online platforms, generating fingerprints where music is detected, and comparing these back to the reference databases has allowed huge volumes of previously unreported music use to be detected, reported and monetised, unlocking sizeable income for rightsholders.
With the truly colossal quantity of content on these online platforms, automated MRT-based solutions are the only viable approach to catching the uses of music that get through the content-checking routines that are applied by the platform operators at point of upload.
But online is not the only domain in which volume and content diversity is a challenge. A major source of income for rights holders, and particularly performers, is where music is used in a public performance setting such as in shops, bars, gyms and nightclubs.
PPL licences around 440,000 such businesses in the UK. Some of the bigger licensees are able to report the actual music that they play. However, it is not reasonable to expect most of these types of businesses to maintain or submit records of each piece of music they play.
In the absence of such reporting, for many collective management organisations (CMOs) such as PPL, the traditional approach to allocating the royalties arising from music used in these venue types is to use proxies such as what gets played on popular radio stations.
On initial inspection, this might seem like a rather crude approach. However, our experience of using this approach, and it is one we regularly reassess and refine, tells us that, when combined with market research and appropriate sampling, it can deliver a high level of accuracy. It is, nevertheless, a further opportunity for MRT to be applied and it is one that is being pushed hard.
Building on the rise of low-cost computing afforded by technologies such as Raspberry Pi and Arduino, where entire computers can be constructed for well under £50, the approach to MRT has moved on.
The key sales pitch of MRT in the public performance domain is one of accuracy for low cost. Why use proxies when you can get accurate and specific data on a venue-by-venue basis for a relatively low cost? But this is where I think we really need to stop and challenge things.
Let’s start by looking at accuracy. There are many elements to the accuracy of what gets reported by an MRT device. Firstly, did the device really hear all the music that was played and was the fingerprint that was generated an accurate reflection of it?
Public performance locations can be very noisy places – especially bars and clubs. Catching a usable feed of the music being played can be challenging.
“Claims of improved accuracy, greater fairness and reduced cost should not be taken at face value, but proven through trials and parallel running.”
Related to this, is a question of whether the fingerprinting and matching algorithm are robust enough to get the correct details from the reference database. In trials we have undertaken, this is not always the case, and it is not uncommon to have music incorrectly reported as a cover version or a derivative recording that actually only contains samples of what has been reported.
Assuming that the MRT has properly detected and fingerprinted what it hears, the second question becomes one of how authoritative is the data that is returned from the reference database. This is of particular concern when users of MRT rely on this data to determine who the rights owner of the recording is.
Many of the major MRT vendors have databases of 120million+ sound recordings. That’s a truly huge number. Managing metadata across a dataset that large is a mammoth undertaking, particularly when it comes to data about rights ownership.
The industry is starting to make inroads into cleansing ownership data through initiatives such as RDx, but it will take considerable time before all conflicting rights ownership claims have surfaced and been resolved. Knowing the lineage of an MRT provider’s data, and knowing what level of trust you can place in the data you get back, is critical.
As a final point on accuracy, even if the music is correctly recognised, there is an important question as to whether it should be included in what is reported. Whereas it is possible to discern from a licensee-provided playlist how a piece of music was used (as an interstitial, in an advert etc.), MRT is blind to this. Its results are devoid of context, so it recognises and reports all music that it hears, including music that may not be covered by the licence.
None of these questions need necessarily be barriers to adoption of MRT technology for public performance music reporting. The point I am making is that you really need to understand the problem you are trying to solve, and you need to properly test the assertions made by the provider to get comfortable that you are indeed getting the accuracy you seek for the cost that is involved.
When deployed at large scale, and there are plenty calling for this technology to be rolled out to tens and possibly hundreds of thousands of public performance venues, the costs can easily run into the millions per year.
While PPL has generally seen good results when using MRT in the public performance domain, and it does form a part of our processes, it has its blind spots.
In trials that we have conducted, we have left devices running overnight in our own office whilst we played out non-mainstream genre music. In some cases, detection rates have been low enough to question the accuracy of the assertions made about accuracy.
A common part of the MRT sales pitch is that when using traditional approaches to distributing public performance royalties it is the niche genres that lose out, and that only MRT can fix this. Our own trials suggest this not wholly to be the case, as it was often those niche genres that suffered from lower match rates. This doesn’t undermine the potential of MRT, just that it drives a need for careful consideration and to use it proportionately to augment other approaches.
MRT has undoubtedly already added significant value to rightsholders, performers, licensees and administrators such as PPL. It also has the potential to deliver even further value with services that are emerging.
But as is true with so many applications of technology, there is a real need to look behind the marketing hype and to truly understand what is being offered. All services are not equal, and important questions must be asked to determine the veracity of the marketing claims.
Claims of improved accuracy, greater fairness and reduced cost should not be taken at face value, but proven through trials and parallel running.
The urge to embrace new technology with bold messages can be high, but as dull as this might sound, we must stick to the basics and do proper due diligence and develop fully thought through business cases.
Doing this not only ensures the best overall outcome for the music industry but ensures the right level of engagement between all stakeholders and helps drive this interesting sector forward.
Music Business Worldwide
