Person attaching modern wireless lavalier microphone with magnet to pink t-shirt for interview or broadcast, preparing audio equipment for recording.
getty
Right now there is a video file sitting somewhere in the cloud that a large language model (LLM) cannot read.
It isn’t because the file is encrypted or corrupted. It’s because raw MOV, MP3 or WAV files are illegible to AI systems in a technical sense. There is not any text for the model to parse and no structure for it to assign meaning. “It cannot read it,” says Owen Grover, general manager of the US market at Nomono, an audio hardware and software company. “You can’t put an MP3 into an LLM.” Before an AI system can search, sort, license or repackage audio or video, something has to wrap that file in data the model can actually understand: a transcript, a speaker ID, a timestamp, something.
Initially, that sounds like a niche technical detail. It isn’t. As media companies lean harder on AI to catalog, promote and monetize the hundreds of thousands of hours of content sitting in their libraries, an uncomfortable realization is beginning to surface: most of that content was never built to be ingested by anything other than a human ear. Nomono is betting that the next competitive advantage in audio and video isn’t better content; it’s content that AI systems can actually “see.”
Podcasts And The Convergence Problem Nobody Planned For
Podcasting is in the middle of an identity crisis. With the rapid proliferation of video podcasts, what used to be a purely audio medium is becoming a hybrid one. Global ad revenue for podcasts and video podcasts is projected to reach roughly $5 billion in 2026. That’s a nearly 20% increase over the prior year, according to Deloitte. Much of that growth is coming from video’s ability to drive discovery through social clips. More than 550 million people now listen to podcasts monthly worldwide, a figure that increasingly undercounts the real audience, since a growing share of that consumption happens by watching on platforms like YouTube, not just listening.
In 2024, Google shuttered its Google Podcast service in favor of rolling podcasts into the YouTube ecosystem. The shift has created friction most creators still haven’t solved. Video-first platforms have become very good at capturing sharp visuals, but left audio as an afterthought. Even 4K footage often comes paired with audio that sounds like it was recorded from ten feet away (because it usually was). Grover points to Apple’s rollout of video podcasting inside Apple Podcasts as a case study in how disjointed this has become. A video episode and its audio-only RSS feed version are technically two separate assets, mixed and mastered separately, even though they’re the same show. There is a similar situation at Spotify, although they seem to be separating the two intentionally to allow the audience to choose their preferred experience: video or audio only.
Where Does Nomono Fit In?
Nomono’s answer is a small piece of hardware called the Sound Capsule. It’s an eight-microphone recorder that also functions as the control center for a recording session. It can be paired with up to four wireless lavalier/lapel microphones that store roughly 45 minutes of backup audio onboard in case a connection drops. Everything captured in the recording gets pushed automatically to Nomono Studio Cloud, a browser and mobile app where an AI enhancement suite cleans up the recording. The enhancement suite removes crosstalk, denoises and balances voice levels across speakers within seconds rather than the hours a human audio engineer would typically spend working soundboard faders by hand.
Nomono also has a lighter version called the Stellar Kit that strips the system down to a single lavalier microphone and a mobile app. The Stellar Kit is aimed at journalists, documentary crews and social video creators who need broadcast-quality sound without hauling a mixing board into the field. Grover demonstrated this with a clip shot on a busy street in Bangkok, Thailand. A Nomono staffer’s partner filmed him on an iPhone from several feet away, capturing mostly traffic noise and market chatter. Once the phone’s native audio was passed through Nomono’s system through the enhanced lavalier recording, his voice came through clearly while the ambient street sound stayed present but no longer overwhelming. This is the kind of result that would typically require a sound editor, not a software button. Notably, the video itself still processes locally on the device rather than uploading to Nomono’s servers. Grover says that was a deliberate choice built around speed and the fact that most creators already have their footage sitting in a camera roll.
Sound Punishes More Than Video Ever Does
There’s real science behind why this matters commercially. In a widely cited study published in the journal Science Communication, researchers Eryn Newman and Norbert Schwarz found that identical academic talks were rated as less interesting, and their speakers as less intelligent and likable, when the audio quality was degraded. This is despite the fact that the video quality never changed. Other communications research has found that audio fidelity has a more pronounced effect on attention and memory than visual fidelity, and that strong audio can even make viewers perceive mediocre visuals as higher quality than they actually are.
The film industry has quietly understood this asymmetry for decades. Anti-piracy engineers have specifically targeted audio, not just video, when designing deterrents against illegal camcorder recordings in theaters. Patented techniques exist that alter a film’s audio track just enough to be undetectable to a paying audience but detectable, and degraded, on a bootlegged copy. In other words, the industry has long treated sound as important enough to sabotage strategically when combatting piracy. Most creators still treat it as an afterthought to fix in post.
The Pitch Against Plugins
Grover is careful to frame Nomono against a crowded field of AI audio tools, from Adobe Podcast and Descript to ElevenLabs, without dismissing them outright. His argument is structural: Those tools are point solutions that clean up audio you’ve already recorded. Nomono is trying to own the entire chain, from the microphone to the published episode, so a creator never has to import “a ton of crosstalk” in the first place. According to Grover, customers save 60% to 70% of the time typically spent between downloading a recording and shipping a finished episode (a claim from the company rather than an independently verified figure). The claim aligns with the broader thesis: The more of the workflow one system owns, the more time it can claim to save.
Betting On The Future
Grover is candid that Nomono doesn’t lead its pitch with the AI-legibility argument. Customers care about clean audio and faster turnaround more. But he calls that metadata layer “almost in some ways the Trojan horse of this product,” because it’s what will let publishers eventually search, license and repurpose enormous audio catalogs using AI rather than combing through raw files by hand.
That’s the more interesting story here, and possibly the more durable one. Better microphones are a commodity market that will keep getting crowded. But an industry racing to make its content legible to the AI systems now doing discovery and licensing on its behalf may find that the real cost of ignoring sound was never a bad listener experience. It’s an asset that no human or machine can find.