Cookie usage policy

The website of the University Carlos III of Madrid use its own cookies and third-party cookies to improve our services by analyzing their browsing habits. By continuing navigation, we understand that it accepts our cookie policy. "Usage rules"

News

  • Home
  • News
  • UC3M software subtitles Spanish Radio and Television (RTVE) regional news reports in real time

UC3M software subtitles Spanish Radio and Television (RTVE) regional news reports in real time

Developed by the SoftLab research group

7/14/20

Software developed by researchers from Universidad Carlos III de Madrid (UC3M) has been implemented in ten of the 17 regional centers of Spanish Radio and Television which can subtitle regional news reports in real time. The system is aimed at improving communication for the approximately four million people who have hearing-related disabilities in Spain.

The UC3M SoftLab research group has implemented a completely autonomous system for automatic subtitling and in real time, through artificial intelligence techniques and machine learning. The software itself transcribes the voice to text and sends it to the system for inserting subtitles that the RTVE has at its headquarters in Torrespaña.  From there, and   simultaneously, information is sent to each of its centers so that every Regional Community can see the corresponding subtitles based on the audiovisual content it is broadcasting.  “The architecture can support generation and transmission of a large number of simultaneous subtitles without affecting the quality or speed of subtitling for each broadcast.  The subtitling process feeds back to the recognizer, so that the more broadcasts generated by the subtitling system, the more accurate future broadcasts will be,” explained Ángel García Crespo, head of SoftLab, professor in the Department of Computer Science, and subdirector of the UC3M "Pedro Juan de Lastanosa" Institute for Technical Development and Promoting Innovation. 

In contrast to the current system of “respeaking”, presently in use and which requires someone to introduce and send the transcription, the system developed by UC3M carries out the process in the cloud, and requires almost no human intervention, infrastructure or specific location.  “The system analyses results in real time from several voice recognizers and constructs a final transcript with results that are superior to individual results from the individual systems,” García Crespo pointed out.  

This software is in accordance with subtitling regulations and produces a maximum of 37 characters in two or three lines, as the norm requires.  Furthermore, it automatically includes punctuation and identifies the correct terms used in each area, which have been added individually by region to obtain a more efficient transcript and reduce the errors produced by other similar systems.