With all the current buzz about artificial intelligence (AI) and machine learning (ML), you might think these tools are new to the media industry. In fact, though we have yet to see much use of generative AI in the supply chain, ML has been a key component of automated workflows for years. And for good reason: years of iteration, training, and learning has made many ML-based tools highly accurate. Today, most modern media supply chains utilize these tools to accelerate the flow of content and make manual operations more productive.
ML in the Supply Chain
Many media organizations have come to rely on ML-driven tools that perform preliminary content review for key compliance markers — alcohol, weapons, nudity, etc. — so that human operators need only review highlighted instances. And today, the latest releases of numerous ML-enabled tools deliver the functionality and performance essential to a wider variety of automated workflows.
Consider, for example, transcription solutions that have improved radically from a not-so-helpful 80% accuracy several years ago to high 90% accuracy now. In many cases, these automated speech-to-text tools supply more accurate results than do traditional subtitling or captioning processes. Having improved significantly from early iterations, these and other ML tools empower media organizations to achieve the efficiency improvements that come with reducing manual touches and refocusing human intervention on other areas of content preparation and delivery.
Modern ML Applications
Modern ML-driven solutions are powerful enablers of efficiency, and that’s why we’re beginning to see them implemented more broadly, with the effect of integrating critical tasks into more streamlined media supply chains. We’ve found that when our own customers deploy ML solutions, they augment human activity in a way that boosts productivity. By integrating automated QC into its ingest workflow, for example, one customer reduced manual touch by 80% and cut the time required to process an hour of content, enabling a single operator to work through one hour of content in just 20 minutes rather than the 120 minutes it took previously. That’s five times as much content, with existing personnel working the usual number of hours.
How else are media organizations taking advantage of ML to accelerate content preparation and delivery? Along with automated QC and transcription, they’re using ML for caption QC to ensure captions and audio align properly. They are using it to accelerate identification of slates, credits, “up next” sequences, and other elements of shows and movies so that content can be marked and subsequently tailored for different viewers. Text and textless detection tools speed identification of on-screen text — whether it’s there or not, what language it’s in, whether it is acceptable by the standards set for the target market. Using pretrained, customizable computer vision capabilities, organizations detect faces, objects, scenes, activities, and more within images and video.
All these ML tools generate immense amounts of time-based metadata as they evaluate content. That metadata becomes valuable beyond a specific use case if it can be captured, normalized, and leveraged as part of a larger data set to provide insights into the content library. With an extensive and continually growing collection of metadata, a media organization can quickly search for and find elements or objects within any of its assets. Moreover, information about all the jobs performed can inform analysis and optimization of content preparation from end to end.
The Rally media supply chain management platform collects metadata from all the tools that run on it, then normalizes that data and collates it back to the assets being processed. Rally ensures that metadata is stored and normalized in such a way that other tools can access and read it, regardless of its origin.
Real-World ML Implementation
The Rally platform gives organizations the ability to try different tools, including ML-driven tools, in their supply chains with very little upfront cost, time, or integration work. They can spin up an instance of a particular tool, run some content through it using a Rally supply chain, and see if it provides the accuracy they require or the metadata guidance that they need. If it does, they can simply integrate it into the supply chain. If it doesn’t, they can fine-tune the tool’s settings or test another solution to find out if it’s a better fit.
In the world of ML and AI models, technology providers play a constant game of leapfrog. One company has a good tool, and then another company improves on it and comes out with a slightly better tool. Rally users benefit from this continual evolution and improvement, as they can easily switch to a new tool that does what they want best. And for any given job, they can apply the best tool for the job and its requirements.
Deployment of ML for content preparation isn’t just a future possibility. Just like editing and localization, it’s now an active part of modern media supply chains. Here’s a look at a few of the ML-based Application Services already available via the Rally platform to achieve immense efficiency gains and greater insight into content libraries:
- Amazon Rekognition
- Amazon Transcribe
- Cinnafilm Tachyon and Wormhole
- DeepVA
- Google Cloud Video Intelligence
- Interra Systems BATON
- Microsoft Azure Video Analyzer
- Prime Image Time Tailor
- SSIMWAVE VOD Monitor
- Telestream Cloud Qualify
- Venera Technologies Quasar
Using ML to go from having operators touch every piece of content to only touching it by exception, media organizations can make the larger shift of processing many times more content without having to add incremental operators. And with this shift, they can realize transformational efficiencies and immense productivity gains across their operations.
—
If you’d like to learn about integrating machine learning-based tools into your supply chain to streamline and accelerate the work you’re doing with your content, give us a call. We’d love to show you how you can do that.
