SI Media has developed its own AI framework with an internal project to start in the second half of 2022.
The aim was to develop a framework, integrated with YES!, to extract descriptive metadata from video content that would be processed and managed by YES!MAM.
The first phase of integration, thanks to our R&D managers, led us to achieve important results with the AI resources we have today:
- The system performs speech to text, converting spoken video recordings into readable text which can be converted into a subtitling file, ready for playout. This function, in recent years, has been upgraded to a higher level with a translation algorithm. This has made it possible to have a complete set of tools, with a dedicated interface, to generate and manage captions and subtitles in multiple languages, recognising the sound source with great speed thanks to automatic speech recognition.
- It also provides visual analysis with the content moderation function, recognizing inappropriate material including explicit visual or prohibited content.
- To further extend the capabilities, we have introduced a brand new object recognition algorithm. This gives broadcasters the flexibility to tailor the set of recognisable objects to their specific needs.
- High priority in the YES!AI update was given to the integration of facial recognition for the automatic identification of people appearing in video content (e.g. leading actors in films). This integration allowed our AI to take a quantum leap forward with the fast recognition of faces in various resources, allowing users to be able to assign names later at their convenience.
The integration with major AI service providers (Google-Cloud, Amazon-AWS, Microsoft-Azure) ensures automatic metadata detection, improving the efficiency and accuracy during ingest and the capability of the advanced search within the Media Asset Management system.
SI Media has also developed a customized feature in cooperation with Google Translation Services to perform automatic translation of metadata inserted by each user in his own language, allowing cross language queries inside YES!MAM. For instance, type a search query in French and get results in English.
The main goal that SI Media wants to achieve with AI is the incorporation of semantic search to transcend the limitations of keyword-based search by digging into the heart of the content: its meaning and context. Let's think about searching for a specific event, not just based on the title, but also on the main themes, the emotions evoked or even the personalities involved.
Thanks to semantic search, AI can unlock this potential by analyzing the entire content, including audio transcriptions, subtitles and even visual elements such as scene recognition.
See the following PDF for more information
