The need
Episodic content is published in large volumes, i.e. tens of hundreds of episodes at once.
The quality of the viewer’s experience requires that content be properly segmented, with accurate markers for intros as well as opening and end credits.
Most importantly, ad breaks must be positioned so that they do not disturb narrative continuity. Such disturbance is highly detrimental to viewer experience, risking viewers leaving the service.
Depending on the type of service (linear, BVOD, AVOD, SVOD …), the applicable regulation framework, the monetization principles and the duration of content, the number of required ad breaks varies.
Given volumes, time and budget constraints, manual placement of markers is not a viable option.
The issue with existing solutions
Some content, e.g. with acts clearly separated by black & silence pauses, makes the detection of break positions easy. However in most cases this is not straightforward.
AI-based solutions have two significant pitfalls: they come up with a high and unpredictable number of ad break candidates; they usually confuse classic editing principles such as shot / reverse shot or very short scene duration editing with narrative changes.
The consequence is that a human operator needs to be involved to make a final selection. In best-case scenarios, the only gain is a productivity gain which (although significant) is not a game changer.
Our approach
Nomalab’s approach combines GenAI, using AWS’ Media2Cloud, with well-proven, computing-economical signal analysis to produce a large set of weighted data points: blacks, silences, fades, scene changes, music segments, dialog-based sequence segments, on-screen text detections, ad break candidates …
Nomalab’s own post-evaluation process assesses each ad break candidate against narrative and editing data points within temporal zones of interest taking into consideration definable rules (e.g. number of required ad breaks, distance, etc.). This post-evaluation emulates the reasoning of an experienced human operator.
The result
The process takes a few minutes.
The result is a unique set of time code markers strictly matching the defined number of required breaks and respecting narrative breaks of the content. These markers are exported e.g. as a JSON set referenced to the content’s media file. They can be ingested and used for publication without any subsequent human intervention.
Markers can be displayed alongside content within Nomalab’s advanced web player.
Nomalab has produced and delivered marker sets for hundreds of episodes for France’s leading BVOD service M6+, with a quality-success rate higher than 90% (a sample of Nomalab’s markers are independently assessed by a seasoned operator).