A Tough Fix
Generative AI has many applications, many of which are beneficial both to individuals and to society in general. Similarly businesses can certainly benefit from AI, although, as we have pointed out, it has a darker side and can be used in nefarious ways that we have yet to fully grasp. One company that has been dealing with the somewhat darker side of generative AI is Spotify (SPOT).
Spotify is based on a number of algorithms. Most folks assume that payouts are made based on the number of times a particular piece of music is streamed by users. This is not quite the case as Spotify uses a ‘streamshare’ system rather than a per play system. Here’s how it works:
Spotify just announced that over the last year it has removed over 75 million songs from their servers. Some were AI generated songs using voices cloned from real artists, others were songs that were Ai generated and uploaded to an artist’s site without permission, and many were the aforementioned ‘slop’ category used to goose numbers. Spotify has been working on a ‘slop’ filter that is able to discern which tunes are intentional duplicates used for stuffing numbers and has instituted a policy that targets AI created music that duplicates artist’s voices without their permission in connection with publishers who can monitor artist sites.
Spotify is also trying to develop attribution standards for audio recordings, essentially a watermark that will indicate whether a song is Ai generated, but if the image and video industries are any example, audio watermarking is next to impossible as there is no visual point where the watermark could be seen. And we doubt consumers would accept an audio disclaimer each time an AI generated or AI-enhanced recording is streamed. It’s a tough world out there and while AI can be used as a creative tool for audio, it has just the characteristics those interested in gaming systems like Spotify need to enhance their income without the creativity that has been the hallmark of the music industry. A tough fix.
[1] In this case a ‘Rights Holder’ is not a single entity but a ‘container’ for a number of people and organizations that participate in administering and financing the creative process. It includes the artist, along with a number of other organizations.
Spotify is based on a number of algorithms. Most folks assume that payouts are made based on the number of times a particular piece of music is streamed by users. This is not quite the case as Spotify uses a ‘streamshare’ system rather than a per play system. Here’s how it works:
- Spotify add both premium subscription fees and advertising revenue from free-tier ads.
- Taxes, processing fees and other operating costs are subtracted from the combined revenue resulting in a Net Revenue from which Spotify takes 30% and allocates 70% to a royalty pool.
- The big recording pool is broken down into two smaller pools, one for Recording Royalties (~80% - 85%) and one for Publishing Royalties (15% - 20%).
- Spotify counts the total number of streams (must be ≥30 seconds to be counted) across the entire platform monthly, on a country by country basis and calculates the ‘Rights holder[1]’ share, essentially the number of streams for that rights holder over the total number of streams, done for each market.
- That share is allocated to the Recording Royalties pool and the Publishing Royalties pools, so if the right’s holder has a 2% streamshare, they get paid 2% of each pool.
- The Recording Royalty streamshare goes to the Rights Holder directly, but that is not the artist, its the record label (if the artist is sighed) or the independent distributor (if unsigned). The record label then distributes the payout to the artist based on their contract with the artist, or to the independent distributor who takes a fee and pays the rest to the artist. Note that signed artists have to recoup the costs associated with making the recording before they get paid. This can include an advance, recording costs, video costs and even tour costs in some cases.
- The Publishing Pool gets allocated using the same streamshare rate, to the song writer (50%) through a performing rights organization like ASCAP, while the publisher gets the remaining 50% but pays the writer a portion of that based on their contract. We note here that the way the Publishing Pool is even more complex and nuanced that we show here.
Spotify just announced that over the last year it has removed over 75 million songs from their servers. Some were AI generated songs using voices cloned from real artists, others were songs that were Ai generated and uploaded to an artist’s site without permission, and many were the aforementioned ‘slop’ category used to goose numbers. Spotify has been working on a ‘slop’ filter that is able to discern which tunes are intentional duplicates used for stuffing numbers and has instituted a policy that targets AI created music that duplicates artist’s voices without their permission in connection with publishers who can monitor artist sites.
Spotify is also trying to develop attribution standards for audio recordings, essentially a watermark that will indicate whether a song is Ai generated, but if the image and video industries are any example, audio watermarking is next to impossible as there is no visual point where the watermark could be seen. And we doubt consumers would accept an audio disclaimer each time an AI generated or AI-enhanced recording is streamed. It’s a tough world out there and while AI can be used as a creative tool for audio, it has just the characteristics those interested in gaming systems like Spotify need to enhance their income without the creativity that has been the hallmark of the music industry. A tough fix.
[1] In this case a ‘Rights Holder’ is not a single entity but a ‘container’ for a number of people and organizations that participate in administering and financing the creative process. It includes the artist, along with a number of other organizations.
RSS Feed