In the last few years, machine learning has made tremendous advances in areas such as computer vision or sound processing. Trained models are capable of identifying patterns and extracting information that very recently only humans were capable of doing reliably.
Images, audio, and video have been pervasive for a long time and media-intensive environments find themselves managing ever-growing archives. With the never-ending expansion of contents in these archives, complexity also increases. When trying to gain new insights and extract new information from this sea of digital content, one would need an army of experts and documentation experts to tackle the challenge.
It's in these situations that the capabilities of artificial intelligence play a very important role, offering automated ways to solve some of these needs. By learning from annotated examples and by seeking patterns in the data, new sorts of information can be extracted. In some cases, this can also be achieved with existing models.
RTS, like many radio and TV institutions, is in this exact situation where they deal with huge archives. As part of their strategy to increase the value of these archives, they have invested in developing a AI platform for their media. One of our data engineers, Ireneu Pla, has been helping them to expand the reach of their platform and performing new analysis.
Working with RTS, Ireneu has been collaborating with a team that has a captivating and ambitious goal: unlocking the contents of their archives. The path chosen to tackle this challenging objective hinges on the promises of artificial intelligence. However, making machines understand content we once thought would be out of their reach is but one of the challenges. «While the technical possibilities are formidable,» Ireneu shares, «the aim of these efforts is to solve a real problem and to put the solution in the hands of users. To make this reality, it takes a healthy dose of software engineering and putting heads together. Finding a suitable machine learning model is just one step.»
Should this make you reconsider your existing investments in the realm of AI? Simone Comte, Product Owner at RTS, doesn’t think so. «Sooner or later investments need to be made to improve aging or outdated systems», explains Simone. «Prior to implementing the AI platform, the identification of people in the archives relied mainly on the memory of a few people, most of whom are retiring. The identification of speakers using AI will help enormously to fill this loss of knowledge without having to hire experts to replace those that have left.»
-
- Simone Comte
Product Owner of RTS-ai
As a result of our work together, RTS’s AI platform now boasts strong audio analysis capabilities, allowing them to shine a light on mountains of various audio content. This new functionality will allow them to classify old records that are being digitized, to collect gender statistics for radio programs, and to improve other processing tasks.
«Another great strength of the platform is the People database. Combining facial recognition, voice recognition and human work, it becomes a unique repository of valuable data», affirms Simone. These capabilities also enrich the metadata in RTS’s Media Asset Management systems (MAM), improving the experience of documentalists, archivists, and journalists.
The Quality Check module is another highlight, as it allows RTS to not only detect errors in the files coming from the digitization process but offers new functionalities that are useful for production, like providing time codes for credits, and potentially for content management such as identifying programs with a jingle.
«The RTS-ai platform will continue evolving, pursuing new ways to analyze and enrich data in order to increase the value of our archives with machine learning», concludes Simone.
Adnovum
