National Media and Infocommunications Authority (of Hungary)

The National Media and Infocommunications Authority is the independent regulatory body of the Hungarian state responsible for overseeing the media and infocommunications market. Its tasks include ensuring interference-free frequency usage, filtering out non-compliant equipment, and supervising the quality of electronic media and infocommunications services. Its goal is to ensure fair competition, media pluralism, and the enforcement of user interests.

The Authority set out to design and implement a system capable of:

• producing textual transcripts of media assets
• identifying speaker changes (identifying speakers)
• recognizing name entities
• determining the start and end of spoken words with hundredth-of-a-second accuracy

In addition, all system functions must be accessible via an easy-to-use web interface.

The above requirements serve multiple purposes. On one hand, video and audio media assets become searchable in text form. Such use cases include content analysis of political news and magazine programs. On the other hand, checks such as monitoring advertisements and sponsorships, tracking product placements, verifying protection of minors, examining age ratings, and supervising accessibility for the hearing and visually impaired can be enabled or automated. Thirdly, in official procedures, it is expected that investigation reports contain verbatim transcripts of the examined program elements.

Until now, transcripts were prepared exclusively by human resources, which was an extremely time- and labor-intensive process. Thus, the Authority did not have adequate IT tools for the efficient monitoring of these tasks and regulations.

A system capable of handling unique and automatic transcription processes, with completed transcripts viewable and exportable (in SRT, Word, and JSON formats) via the web user interface. The system includes an accuracy measurement module that evaluates the quality of machine-generated transcripts using objective indicators (word and character error rates, speaker changes, timestamp and named entity recognition accuracy). The system’s trainable models guarantee consistently accurate transcripts and long-term support for the Authority’s tasks.