Sound designers form an alliance against AI scraping, but their fight is about price and consent, not refusal

Music has labels and the RIAA. Voice has SAG-AFTRA. The footsteps, door slams, rain and engine roars under every film, game and a lot of records have had nobody. On 24 Sep 2026 a group of sound designers, mixers and library companies launched the Professional Sound Alliance to change that, with a slogan reported by Music Business Worldwide: audio has three pillars, and only two are protected.

What is the Professional Sound Alliance?

It is a coalition plus a public petition. The alliance site says it speaks for recordists, designers, editors, mixers, supervisors, composers, students and the studios and libraries behind them, and it lists Pro Sound Effects founder Douglas Price as chair. Founding names include Oscar-winning sound designer Mark Mangini (Mad Max: Fury Road, Dune), Boom Library CEO Pierre Langer, Sound Ideas VP Craig Nimens and re-recording mixer Karol Urban, a former Cinema Audio Society president, per MBW. When we checked on 23 Oct, the site showed more than 900 signatories.

It names two problems:

• Data Piracy: sound effects and sound design pulled into commercial AI models without permission.
• Sync Piracy: sounds, or whole libraries, used in films, shows and games without a licence.

The site calls these different methods of the same theft.

What it is not

It is not a union, and it isn't suing anyone. MBW reports the alliance has no collective bargaining agreement and no litigation fund, and that it says it has no intention of using the group to litigate or to set licensing standards. The petition is described as not a legal document and not an assignment of any rights. For now it is a petition and a downloadable seal; Production Expert describes a coalition campaigning for fair licensing of professional sound.

The founders already sell to AI

This is the detail that reframes the story. The alliance's founding libraries are not refusing AI. They are selling to it.

• Pro Sound Effects markets a training dataset of 1.27 million sounds and more than 5,000 hours, delivered by API or bulk download. It shows Google, Sony AI, Dolby, Samsung and Bose logos under "trusted by" and says it owns 100% of its catalogue, so buyers face no rights disputes. The same page lists its membership of the alliance.
• Sound Ideas offers its catalogue, which it puts at 1.2 million sound effects across more than 400 libraries, for training AI models, with generative audio and text-to-sound-effects among the listed uses.
• Sony AI trained Similar Sound Search, a Wwise feature announced with Audiokinetic in August 2025, on licensed BOOM Library and Pro Sound Effects libraries. In May 2026 it unveiled Woosh, a sound effects foundation model for text-to-audio and video-to-audio. Sony trained a private version on licensed Pro Sound Effects and BOOM libraries and released a non-commercial public version trained on openly available data.

Pro Sound Effects told MBW the complaint is about unauthorised use and scraping, not about generative AI as such. So the alliance's real position is clear: train on our sounds if you like, but ask first and pay.

Why now?

The music side shows what happens without a licence. MBW, citing 404 Media's July report on leaked Suno code, notes that Freesound and Pond5 were named as training data sources. Freesound hosts Creative Commons recordings, and Pond5 is owned by Shutterstock. Sound effects are already inside music models, and nobody negotiated for them.

MBW also points to a cautionary case. Musician Anders Manga sued Stability AI and AudioSparx in December 2025, arguing his recordings were used to train Stable Audio in breach of a licence. In March a judge dismissed the claim against AudioSparx, finding the licence covered the use. Whatever you think of the outcome, the lesson is that the contract you signed years ago may already decide your AI rights.

What it means for recordists and producers

Producers reach for these libraries constantly, for risers, impacts, foley and textures. The people who recorded those sounds are the ones with the least visible stake here.

The alliance site credits the craft well: field recording with specialist gear, up to dozens of hours of work behind a single effect, and human-written metadata that makes a library searchable. That metadata is exactly what makes a catalogue valuable as training data, because AI companies want labelled audio. Pro Sound Effects sells its dataset on the strength of human-written descriptions and a consistent tagging system.

What the AI licensing pages don't say is how that money reaches the recordists. Pro Sound Effects names Mangini, Oppenheimer sound editor Richard King and "20+ other" sound artists on its AI page, and speaks of supporting artist earnings. But it gives no detail on contributor consent for AI use or on revenue share. Sound Ideas' page says nothing about either. The alliance's own launch material doesn't address it either. That gap matters most to the freelance recordist who sold a library outright, or uploaded to a marketplace, before anyone was training models.

What we think

The Professional Sound Alliance is best read as a market-making move dressed as a rights campaign. That isn't a criticism. Library companies with clean, wholly owned catalogues gain the most if "scraped" becomes toxic and "licensed" becomes the norm, because they are the only ones who can sell licensed data at scale. Pro Sound Effects says so on its own page: 100% ownership, no rights disputes. Scraping is their competitor.

That helps sound designers too, since a licensed market beats free extraction. But it repeats a pattern from music. The catalogue owner signs the AI deal, and the person whose work is in the catalogue finds out later. The musicians' union AFM is currently suing Universal, Warner Records and Atlantic over licensing members' recordings to AI companies without paying or crediting them, per MBW. Sound effects have no union to bring that case.

Two things would show the alliance is about more than library pricing. First, a public commitment from member libraries to share AI licensing revenue with contributing recordists, or at least to tell them when their sounds are licensed for training. Second, AI licence clauses that sample-pack makers and smaller libraries can copy. If both happen, the "third pillar" could hand the music side a working template for sample packs, which have the same problem and no organised defender. If not, it will be a well-designed seal on a few good dataset sales.

Our advice to anyone who records or designs sound for a living: re-read your library and marketplace agreements for AI and machine-learning language, sign the petition if you agree with it, and ask your library directly whether your work is in a training dataset and what you're paid for it.