Vicomtech and RTVE perfect automatic subtitling

Vicomtech's Management of Speech and Natural Language Technologies in collaboration with the Accessibility and Accessibility Support Engineering Units of TVE have carried out tests aimed at improving quality and productivity in the field of automatic subtitling, under the framework of the I Call for Impulsa Vision Empresas focused on the Accessibility of Audiovisual Content.

The IDAZLE platform generates subtitles of pre-recorded audiovisual content automatically, based on a mixed approach that combines algorithms and artificial intelligence. Specifically, it is already capable of transcribing the audio of multiple audiovisual content into text, as well as capitalizing, punctuating, and segmenting said transcriptions.

The quality of the subtitles in different TVE programs and the post-editing efficiency of the automatic subtitling have been evaluated. In addition, the tests have served to collect suggestions for improvement made by the TVE Units that have participated.

As a differential fact of the procedure used, it should be noted that 57% of the subtitlers of TVE have collaborated, who have individually post-edited four different audiovisual contents, without prior knowledge of IDAZLE. Such a large population sample is truly exceptional. For this reason, the conclusions obtained respond to the operational reality and allow informed decisions to be made.

The results have been very promising. On the one hand, it has been found that IDAZLE provides a clear and intelligible transcription, placing the average quality evaluated at 4 points out of 5. This implies that, on average, only between 10% and 25% of the text needs to be post-edited transcribed. And, on the other, it has been found that the average subtitling time is practically halved in relation to manual subtitling, thus doubling average productivity.

These values are really high and IDAZLE has potential for improvement ahead, which will come both from the improvement of the platform and from the education and training of its users.

All this has encouraged RTVE and Vicomtech to continue collaborating to achieve a solution that will become a reference within Spanish broadcasting, and, in turn, that is perfectly valid for any entity that needs to subtitle recorded audiovisual content.

VICOMTECH is a technological center for applied research and a reference specialized in Artificial Intelligence, Visual Computing & Interaction. At Vicomtech, technologies are researched and developed in response to market demand and providing innovative solutions to companies.

Impulsa Vision Empresas is RTVE's initiative aimed at consolidated entities that wish to collaborate with RTVE in a practical way, testing their innovative projects and products in the technological field of the audiovisual sector at RTVE's facilities and equipment.