AI-generated music is morphing into a multi-billion-dollar industry that no longer sits on the fringes of the music business.
A recent study commissioned by GEMA and SACEM, conducted by consultancy and research group Goldmedia, projected that the generative AI music market will reach over $3 billion in value by 2028.
It’s no surprise that prominent music companies like Universal Music Group have embraced the creative possibilities presented by ‘ethical’ and legally trained AI models.
Evidence of this can be found in UMG’s recent partnership with startup KLAY on a “commercial ethical foundational model for AI-generated music” and its partnership with YouTube to develop AI music tools.
But beyond headline-grabbing examples of copyright-compliant partnerships with major music companies, the wider generative AI music space still faces significant legal challenges due to a lack of licensing of training data and ethical questions around the emulation of real artists’ and songwriters’ work.
The music industry is keeping a close eye on recent lawsuits from labels who argue that AI companies are using their music without permission or compensation, like the ongoing litigation against two of the most controversial players in the entire music industry right now, Suno and Udio. Both companies have as much as admitted via court filings that they used copyrighted recordings from the companies that sued them.
To legally use copyrighted music for training, AI startups need extensive licensing agreements with both recorded music and publishing rightsholders, but the absence of a standard licensing framework for training AI models makes that a complex obstacle to navigate.
Berlin-based independent music distributor Paradise Worldwide wants to change that – and it believes that monetizing AI training data will become an important revenue stream for independent artists and songwriters.
The company, led by founder and MD Ralph Boge, advocates for simplifying access to legal, rights-cleared training data and models to ensure that ethical AI music generation platforms are copyright law compliant.
“Gen AI platforms are mostly illegal,” says Boge. “But they need to become legal with our help, and we need to explain the rights acquisition and technical process to the music market and everybody else that’s a bit nervous about this AI thing.”
Paradise started as a distributor and services company for independent artists and labels in 2009. It has since added publishing administration to its services, including direct negotiations with collecting societies and DSPs. It also collects Neighboring Rights income worldwide and promotes content on DSPs via its in-house agency.
Today, the company claims to be the “only Independent All Rights Distributor” with its own AI department (working across training data as well as Gen AI production and releases / output). Geographically, Paradise is focused on Europe, Africa, and the Americas, with offices in Berlin, Johannesburg, New York, and Mexico.
According to Boge, the industry’s key challenge in implementing new licensing models for AI-generated content is the “rights split” between recording and publishing rights, which, he says, “makes licensing and distribution of revenues complicated.”
He claims that his company’s recently developed All Rights model, which allows it to manage recording and publishing rights on behalf of its clients and capture the extended metadata required to license music for use by Generative AI platforms, is the solution.
“When you get the content from the client, there’s a front-end that requires a bit more data fields than the industry [commonly uses] now. The development of extended metadata, including complete publishing metadata, is quite important.”
He adds: “If the IPI [publicly accessible international identification number], the ISWC [International Standard Musical Work Code] and correct Writer Splits, Publisher and CMO / IME data (if available) – I call it the complete code set – are missing, rights collectors on the publishing side are not able to get the complete revenue [owed to them]. It’s important that the industry works on this, and that’s something we have done [via the All Rights model]”
Parade Worldwide’s All Rights framework forms the basis of a new venture called AIxchange, which is dedicated to licensing AI training data to Gen AI platforms.
AIxchange offers two separate products, which it claims will “enable any GenAI music venture to achieve legal compliance”.
The first of those products is a Large Music Model, which it says is already trained on its own curated training data sets.
The AIxchange venture also provides access to that rights-cleared training data set for AI companies to legally train their own models. The company’s legal training data has already been used in research conducted by Germany’s Fraunhofer IDMT Institute.
Boege says: “Our Gen AI project is called Paradise AI Kitchen and we started to release music in October. It is curated by humans and generated by [AI]. With the help of the Fraunhofer IDMT Institute, we were able to develop a system, which helps us better understand the AI licensing process but also enables our clients to use AI as a creative tool.”
Here, Boge tells us more about Paradise Worldwide’s positioning in the market, his ambitions for the monetization of training data, and the expansion of the legal generative music AI sector…
Tell us about your broader activities in the industry leading up to the development of the All Rights model?
We are active in communities such as the CMO group under the AFEM [the Association for Electronic Music] in order to better understand the CMOs in relation to DSPs and PROs [in relation to] monitoring performance plays. We are currently working on a new AI format / group, integrating the needs of the industry and defining a licensing model that will support the writer better than before and will be a role model for DSPs and Gen AI platforms.
We collect Mechanicals from DSPs where possible, which is a much more efficient system than via CMOs. Our extended Publishing (All Rights) setup helps us identify the owners of the respective rights.
We started delivering all our clients’ data to relevant Music Recognition Tech (MRT) services years ago, when almost nobody believed in these services. Nowadays even Germany’s GEMA uses] a recognition service called Soundaware, which is great.
MRT is not just good for Performance, it’s also a benefit for CMOs such as APRA AMCOS. We are quite close to DJMonitor, the leader of the festival monitoring industry [and have benefitted from] lots of knowledge exchange during the years.
Also to mention is seeqnc, who are getting into the more granular MRT process to better identify AI results.
[Our] experience in MRT services helps to develop AI Recognition [technology]. After we set up the legal and production side of Gen AI, [MRT] is the task to be developed.
We often talk at conferences about the combination of AllRights and the necessary tech [capabilities] of a Distributor in order to have control and to pass on better data to the exploitation industry.
What is the All Rights model and why did you decide to develop it?
This enables us to distribute and collect not just Mechanicals but also Neighboring Rights (within the Distribution service) – and it provides the basis for AI Training Data and our own Gen AI.
In our eyes, it is crucial to be able to handle both rights. First of all, you need to have the rights to produce GenAI [content] and to distribute the rights accordingly.
We advocate for a 50/50 split between Master/Copyright and no random payments to CMOs who can’t handle the distribution of revenues due to a lack of complete data.
Second, it is the way forward to a more transparent ecosystem.
Imagine if the CMOs [received] a percentage of all GEN AI productions, without knowing [whose rights] are inside in the data. This would bring us back to the times where CMOs had a system we called the Blackbox (with lots of unidentified content / metadata and unpaid revenue). [This is] good for the Top 10 en artists, [but] a disadvantage for the Independents [due to uncollected cash returning to the music industry’s biggest players by way of market-share-based blind payouts].
All Rights will also help the DSPs solve their Mechanical issues. Lets be honest, distributors have no interest in complete publishing data, it is intense and costs money [to developmen] [which means that] the DSPs accordingly have the best excuses to not be able to pay the right amount and include the right data.
How is the company working to build bridges with collecting societies and implement new licensing models for AI-generated content?
We have good relations with CMOs. Some of them understand that the timing is right for them to become more valuable if they have more precise and granular tech and that AI is a big chance for the industry to change things for the better. Apart from that it generates new Mechanicals, so why should they not be keen on playing a role?
What steps should the industry be taking to address the lack of proper metadata in the music industry?
In my eyes Master/ Recording only models do not make sense, you have to know about Publishing data and enrich that with the Copyright Supply Chain Services. You become an IME and useful for the Copyright industry.
What changes would you like to see in the industry to ensure fair remuneration for artists whose content is used in AI training data?
Licensing skills considering the ecosystem of the music industry
- All Rights data complete on the Distributor and CMO side, supporting the PROs who still have a lack of tech knowledge
- Notification in case of usage
- A 50/50 split between both rights (Gen AI is replacing the Performer part and the Writer should have more value)
[In terms of] tech requirements/rules for Distributors and CMOs: to be able to handle all splits and to have a common understanding of the industry in not taking random percentages for AI use without having the skills to distribute it accordingly.
How receptive are you seeing Gen AI companies to these conversations?
Our experience is that training-data licensing is something the Gen AI companies are afraid of due to costs. I think they try to work on arrangements with the majors [record companies] or CMOs to solve the “issue”.
This is a disadvantage for indies and songwriters, probably also for [major] labels.
So far, the majors [have been] fast [to innovate in this space] and the indies are not really on track.
Can you highlight the challenges of identifying and licensing content used by Gen AI platforms?
I think the industry has just started to understand that Gen AI platforms need to be monetized. This requires data, knowledge and tech from the independent aggregators.
The Gen AI platforms need to understand that legal content makes more sense to grow. It’s the same situation we had many years ago with YouTube. It took a while to get there, [but now] the industry and the platform have a great relationship and can generate revenue without having sleepless nights.
I can’t imagine that investors in Gen AI models like the situation [regarding] all of these [copyright infringement] cases [against AI companies].
We have invested a lot in our AI ecosystem. To make this happen, you need an experienced tech team, an idea of licensing, legal expertise, etc.
Germany is a complicated place [in terms of] VC [funding], but with the help of science institutes like Fraunhofer and steady All Rights development, we have reached a point where we can say that we found a way to handle AI-generated content.
Music Recognition Tech Services are not there [yet]. Identifying all Gen AI parts based on data requires more than just doing fingerprints or stems and comparing it with the result. We [might] venture [further] into AI / MRT, if we find the right (financial) partners.
Tell us about the importance of capturing extended metadata in the music industry and the challenges to acquiring content with complete metadata?
Comparing both (Master/Publishing) rights is based on complete metadata. It took us a while to develop the front-end publishing (dashboard) for this.
When you act as a Distributor with extended publishing metadata, you can kill two birds with one stone:
You can collect all rights (from DSPs or CMOs/PROs), and you deliver better data to the industry in general. You can also clear AI training data, which leads to Gen AI production/ exploitation.
Knowledge about data helps to develop new formats and control the [AI] machine and not the other way around.
What are your predictions for the industry’s relationship with Generative AI platforms from a creative, licensing, and remuneration perspective in the coming years?
More and more of our clients use our Gen AI for creativity input, to develop their [primary] genre in a new direction, and to have stems for collaborations with other artists.
We hope that the industry will advocate for the licensing [of AI] based on training data approval by rights owners and a complete rights overview on both rights.
We hope that a 50/50 split between Master and Publishing is the way to go. Why? Gen AI replaces the production and passes on more shares to the writer.
I do not think that the DSPs will be able to adjust their systems that fast, toward a better Writer share in the case of AI content. The Gen AI platforms will probably implement [that] before [the DSPs].Music Business Worldwide