NAB 2025: AI and the Three Bears
It's surely no surprise that NAB would have a lot of companies talking about including AI or Gen AI in the latest product releases. The challenge is to find companies which fit. Just like with Golidlocks and the Three Bears, when it comes to how AI is changing the nature of media technology and consequently the media workflow there tends to be three categories:
- there's absolutely too much use of the 'idea' of AI with many vendors at the conference,
- then there's the 'too little AI' camp. This isn't a measure of their AI but rather the functionality it can return and also the problem it's solving
- and finally there are the gems which have used this tech in just the right way.
Many vendors had metadata products which extracted and normalized the metadata. These products began to blur together because their pitches sounded very similar. Without customer references, they are too complicated for the time being to cover quickly.
Then there’s the ‘too little AI’ camp. One media executive said he saw a demo which was reliant on external connections to secondary vendors. The problem with this is you’re too far removed from having certainty about how a feature will work, because the initial developer is married to a second developer with whom they may or may not play nice (think something like a social media company changing their algorithm).
Finally, there are the gems which have used this tech in just the right way. Audio tools were again a big part of very approachable use of AI and Generative AI. Post-production and compliance also proved additional good use cases.
Another industry analyst said that the choices you make now about how you can on-board the data within your content or back-office customer information or operations will really determine the direction your company will go with their AI strategy. If this is true, what makes the most sense is to find part of the workflow where there is a very discrete task which can then be incorporated into your existing vendor’s environment. AI Interoperability needs to be a goal.
Here are the most noteworthy implementations I saw at NAB this year,
Audio
This company provides real-time AI translation using a cloned voice for existing commentators in over 120 languages for live and VOD content. They can distinquish between multiple speakers. They promise sub-3-second latency for creation of both audio and 708 captions. Their product is a no-code solution which accepts live feeds in RTMP, HLS, SRT, and MP4. They promise a 97% accuracy rating (Bleu metric), support regulatory compliance, and currently partner with a number of encoder companies. They recently closed a $14 million series A financing.
Speechmatics
While the previous audio translation and dubbing company works with customer’s live content, Speechmatics provides API, SDKs, and even their own UI to try out their technology. They pitch their special sauce as automatic speech recognition (ASR) software that helps especially with accents in spoken words. They promise sub-1-second latency and can transcribe 40+ spoken languages into text. This is built on self-supervised machine learning. Their engine adapts to different accents, dialects, and environments, making it an enterprises needing scalable transcription. They are used across multiple sectors, including M & E.
Audioshake
Audioshake fixes the need to extricate and isolate specific soundtracks. Think isolating clean speech from noisy backgrounds, several people speaking over each other, removing copyrighted music to avoid takedown or license requirements, music over narration, or music over sound effects. They started off as a music industry application for using AI to separate audio tracks, like vocals, drums, guitar, etc., allowing remixes, karaoke, localization, or sync licensing. Now the tool has found a home for anyone who needs to dissect their audio. In a full room of software demos, they gave me headphones to hear what they did and it was an incredibly welcome experience after the barrage of software interfaces most others showed me. Available via their platform, API, or SDK.
Asset Creation, Verification, and Management
Flomenco
This company says they have a MAM 2.0 product. Their system indexes video assets to create unified data sources. The resulting embeddings can then be used within their workflow automation services. They connect to your storage and provide a rich look into the kind of information you'll need from your assets to do anything further up the AI path. This is a no-code, drag-and-drop environment automating media supply chain workflows with pre-built connectors to products like Ateliere, Fabric Studio, and Rightsline. The company won the NAB 2025 Pilot Innovation Challenge. Their blog has a number of interesting use cases from Salesforce integrations to working with Google Sheets and Fabric Studio.
Postud.io
Postud.io is collaboration cloud-based platform for post-production. They have a pay-per-use environment where production teams can remotely work using either many name brand applications used for video editing, VFX, color grading, sound production, and other post-production requirements. Think Premiere Pro, DaVinchi Resolve, ProTools, After Effects, CInema 4D, Photoshop. An affordable pay-as-you-go service for either these name brand applications or their own more affordable products for lower-budget productions. Customers choose their CPU, RAM and graphic card preferences and are billed per use. They've been in business for three years.
SnapStream
SnapStream is a DVR and clipping platform for broadcast TV content. They monitor live TV with AI and allow for creation of clips with live side-by-side transcription. Users can search TV recordings via indexed closed captions to create real-time clips for broadcast delivery as well as social media. They have cloud and on-prem options. They work with many media brands; NBC, CBS, ABC, CNN, The Daily Show (and the other original spinoffs), Sinclair, PBS News Hour, NPR, and a lot of other well-known media, entertainment, and sports brands.
OpenOrigins
OpenOrigins helps media companies fight deepfakes by ensuring content has verifiable provenance. Their technology uses watermarking to the asset’s path from creation to consumption. By embedding metadata and establishing a trusted chain of custody, OpenOrigins gives creators and publishers the tools to maintain chain-of-custody of assets and assure audiences of content integrity. They have worked with Independent Television News (ITN) to validate every item in ITN’s archive. They work using the blockchain, but frankly, many people in this space spend a bit too much time talking about how the blockchain works and too little time talking about the outcome. You don’t need to know how your recommendation engine works exactly, and the same thing applies here to their blockchain tech.
Related Articles
One of the ongoing complaints about the ad industry is a lack of transparency about where you're buying and how it's measured. At CES 2025, Disney discussed their new global ad tech stack and how they're attempting to solve this problem and bring an easier and more understandable approach to ad buying.
10 Jan 2025