Nomono Labs Opens the Engine Behind Nomono Studio to Developers, Producers, and AI Agents

Nomono, the Norwegian audio technology company behind the Sound Capsule and Nomono Studio, today launched Nomono Labs. Labs makes the speech-processing technology behind Nomono Studio available to developers, technology partners, and production companies through an API and an MCP server. The first release targets immersive and broadcast production. Producers upload the lavalier tracks and the ambisonics recording from a shoot, and each speaker comes back separated, enhanced, and placed where they stood in the room, as a binaural mix or an ADM master in the Dolby Atmos, EBU, and IAMF profiles.

Immersive production comes first because it is the hardest spoken-audio problem there is: several people, moving, in real rooms and locations, kept in sync with cameras. A pipeline that solves it also handles podcasts and interviews, and it produces what voice AI lacks today: real speech in real rooms, with a record of who spoke, when, and where.

Nomono will demonstrate the workflow live today at Immersive Warsaw, in a Dolby Atmos cinema, for an audience of immersive creators, production companies, and technology vendors. An alpha program with a hand-picked group of immersive producers, educators, and software makers is already under way.

Video caught up. Audio is now the bottleneck.

For a decade, immersive production has been driven by the picture: better and cheaper 360 and spatial cameras, mature codecs, and editing in tools like DaVinci Resolve. Audio integration has not kept pace. A typical shoot records an ambisonics sound field plus a lavalier mic on each person, but none of those files say anything about positions or audio scene structure.

Editors cut on flat, essentially mono audio. Sound engineers then often spend two days per production placing, denoising, and leveling every voice by hand, and redo parts of the spatial mix whenever the picture edit changes. Video-based tracking, the only semi-automatic option, is often slower than working by hand and adds a heavy CPU load to 8K and 16K projects.

Listeners notice: when a voice sits in the wrong place, or the reverb doesn’t match the room on screen, viewers sense something is off and immersion is broken, even if they don’t realize what made them drop out of the story.

Automating the realism layer

Nomono Labs does this groundwork automatically, so engineers can spend their budget on story and emotion. Five products (Separation, Enhancement, Spatialization, Analysis, and Delivery) run on the same processing chain already behind every Nomono Studio recording. They isolate and clean each voice to broadcast loudness standards, measure each speaker’s position from the ambisonics recording and follow them as they move, and report who speaks when. Outputs include processed and unprocessed files per mic, stereo and binaural mixes, and spatial audio master files that open directly in video editing tools like DaVinci Resolve or DAWs like Pro Tools.

Labs works with recordings from third-party gear, not just Nomono’s own Sound Capsule. At launch it accepts any lavalier kit alongside a Sennheiser AMBEO or Zoom H3-VR ambisonics microphone, with more arrays to follow. No other service automatically tracks speakers in a sound field, and Nomono holds granted patents on both known methods for doing so.

The chain is proven on real productions, from a Nomono recording to EBU ADM, through the Orbit authoring tool, to IAMF on the Esfera platform. Because positions arrive as standard ADM metadata, they also drive sound design: Nomono’s ADM output works seamlessly with tools like Sound Particles, turning a day of manual foley work into minutes.

No audio pipeline required

With Labs, developers and AI agents get this processing without building an audio pipeline of their own. Developers send recordings to a plain HTTPS and JSON API and get finished files back. The MCP server gives the same capabilities to AI agents such as Claude, ChatGPT, and Cursor, so anyone can ask an agent to clean up a full shoot, deliver a track per speaker, or prepare a transcription-ready file, one recording or a whole batch at a time.

“We built this chain for our own hardware and run every Studio recording through it,” said Jimmy Brodd, VP Engineering, Nomono. “Opening it up means any recorder, any lav kit, and any AI agent can use the same engine, and every improvement reaches all of them at once.”

Teaching machines to hear the real world

The same engines will extend to post-production tools, archives, and transcription and voice-AI products. The larger opportunity is machine hearing. Most speech models are trained on synthetic data: clean speech with simulated noise and rooms added on top. Real conversations happen in real places, with reflections, overlap, and movement that simulations only approximate, which is why a voice assistant can still mishear someone on a busy street.

Nomono takes a deliberately contrarian position: models built on real-world speech in real-world places will outperform models built on approximations. Its technology captures not just a voice but its room, who spoke when, and where. That structured picture helps devices and AI understand who is speaking and where, with applications from healthcare and robotics to the smart home.

“The industry is betting that models will learn to hear through bad audio on their own,” said Kristine Snyder, CEO of Nomono. “We think the better bet is to understand real voices in real rooms. Immersive production is where we prove it.”

Availability

Nomono Labs is available now as an alpha program for a limited group of partners. Each partner is onboarded by hand, and processing is free during the alpha. Pricing will be usage-based, per processed minute and per product, and will be set with alpha partners before commercial launch. Producers and developers can visit http://www.nomono.co/platform-alpha to learn more. API and MCP documentation is now available at https://platform.nomono.co/docs/api/introduction/. Next on the roadmap: per-product jobs, Eclipsa and Apple ASAF export, more supported microphones, and expanded metadata.