IBC 2017 Preview – Artificial Intelligence (AI) and Machine Learning (ML) in Media

Subscribe To Download This Insight

3Q 2017 | IN-4730

The International Broadcaster Conference runs from September 15th to 19th. One of the most significant new trends promoted at this conference will be related to the implementation of artificial intelligence (AI) and machine learning (ML) in video services. Some solutions targeted as AI or Machine Learning simply migrate from editor- or developer-coded optimization methods to neural-network trained solutions, a host of new solutions leverage video analytics to generate metadata.

Registered users can unlock up to five pieces of premium content each month.

Log in or register to unlock this Insight.


Impact of AI & ML on Media Companies


Artificial Intelligence (AI) & Machine Learning (ML) have a number of key functions for media organizations. 

Optimize, Optimize, Optimize.  Media organizations are under pressure to optimize business outcomes, resource utilization and streamline the organizations.   

  • Optimizing resource utilization (storage, delivery, etc) can help to proactively build systems to manage asset storage, propagation, delivery and quality of experience (QoE). Similar to solutions described in optimizing business outcomes, the ability to develop optimization functions and automatically adjust to them can significantly improve operational outcomes as well as service costs.
  • Optimizing the organization occurs through automation of workflow, use of cognitive AI / ML to replace or assist human resources in time-consuming tasks, such as sub-titling, clip generation, translation, and ratings, library metadata creation projects, and other similar functions.
  • Optimizing business outcome occurs by replacing manually defined functions with AI-driven optimization functions. These functions can find true maximizations, while typical editors or experts will be biased by their experience, their inability to define an equation based on the full number of variables that impact a decision, or their inability to segment the conditions which matter, creating separate optimization functions for each of these. While AI solutions can act as a bit of a black box, they do have the ability to optimize business outcomes, and respond faster to changes in the optimization criteria compared to manual process.

Simple, Intermediate and Complex Scenes. While there is a data science skillset which is challenging most media organizations, one of the very interesting things in media is that solutions span from “one dimensional” constructs, such as analyzing metadata or text to “two dimensional” including speech or images up to “three dimensional”, including video analysis.  Solutions will take some intermediate forms for computational efficiency.  For example, a solution could analyze one still frame per second a movie to decrease to a “2.5 dimensional” representation from a 3 dimensional representation.  Each dimensionality significantly increases the processing time, as well as data storage requirements.  Understanding the dimensionality of the solution, as well as cost per dimension, is critical.  

Solution Roundup at IBC


Some of the solutions already announced, or which ABI has been briefed on include.

In the video analysis and categorization space, tapping into AI/ML via SaaS or Cloud-based solutions:

  • Start-up Dimensional Mechanics has a cloud-based programming platform based on a new scripting language (NeoPulse Modelling Language, or NML) which is designed to enable users who are not data scientists to implement AI & ML. Neural networks can be trained and executed in the cloud, and is released into the AWS marketplace. In the media space, Dimensional Mechanics has solved a variety of problems including detecting adult content, and identifying news sources to help and manage against fake news. Dimensional Mechanics released its product into general release in June, 2017 and has closed a total of about US$7.7 million dollars through seed rounds and a series A.
  • Valossa is a start-up, spun-off from the University of Oulu, announcing a US$2 million round of funding at IBC. Valossa has a pre-trained, turnkey SaaS platform, launched at NAB in March of 2017, which can appropriately handle audio, video and textual metadata to categorize video, with easy to use data analysis tools. It has trained its video on a robust set of metadata about 140K movies (made public through the whatismymovie portal), as well as feeding about 9 million visual frames through the tool. Valossa has experience in looking for profane or explicit content, identifying celebrities, and information analysis within the corporate context.
  • IBM Watson’s Media team announced a video enrichment service, leveraged by the US Open as well as several other high-profile customers. Similar to other platforms, the Watson platform taps into text, audio, image and video data to generate a variety of insights. Modules proven in the Watson system include content search and discovery (enriched content, generating more types of metadata in a more automated fashion), recommendation uplifts, closed captioning in real time, highlight clipping – tapping into clues to automatically find exciting scenes and scene boundaries, compliance (including adult content, ratings, etc), and logo identification for branding, advertising substitution, product placement, etc. IBM Watson has worked with TED on the catalog discovery, the Masters Tournament and the US Open.
  • Google Video Intelligence (GVI), an API based in the Google Cloud, helps to automatically generate metadata for large libraries; media asset management (MAM) company Cantemo is tapping into Google Video Intelligence to simplify metadata creation using GVI.
  • Microsoft Azure rebranded its video breakdown tool to Video Indexer – it provides a host of AI functions including transcripts, face detection, speaker indexing, visual text extraction, scene detection, translation, key word extraction, detection of explicit content, etc. Video Indexer is a bit more of a “DIY” solution rather than a full media solution, however, the functionality is available in a more flexible environment in Microsoft Cognitive Services in the Azure cloud.

What is the AI / ML Opportunity?



Failure to adopt AI / ML in video solutions is a survival game.  Software and cloud-based solution vendors serving the video sector that do not adopt AI / ML into their products will lose competitiveness to those that do.  For the benefits described in section one, benefits of outcome, resource utilization, and organizational efficiency, non-AI/ML based solutions cannot remain competitive from a design win standpoint beyond the 2-5 year horizon. 

However, there are also new opportunities created by AI / ML.  Today, companies are talking about US$0.20-$0.30 per minute of video for AI / analysis of video data.  Large consumers of AI / ML services can of course pay less (including, for example, 24 x 7 broadcast solutions or large media library tagging projects).  The amount of data generated by media, news organizations and enterprises – not to mention user generated content – is large.  While not every application merits this level of investment, as prices fall video tagging will become as ubiquitous as photo tagging on Facebook today.

In section one, we discussed three optimization problems where AI & ML will succeed in media operations.  However, the scope of business problems where AI can be applied are, of course, more complex and nuanced than that.  From consumer relations, regulatory, advertising optimization, consumption optimization, catalog management, editorial, etc., having a robust set of tools is one opportunity while applying it to specific business challenges is another opportunity.  We’ll no doubt see specialists in both approaches.


Companies Mentioned