Artificial intelligence becomes a greater prominence in the audiovisual sector
Telefónica Audiovisual Services and Etiqmedia analyze in a webinar the advantages that artificial intelligence in speech technologies, image processing and processing of natural language in the Broadcast and institutional environment.
Telephone Audiovisual Services next to Etiqmedia He organized last Thursday, March 18, a webinar in which the possibilities opened by Lto artificial intelligence (AI) in the audiovisual sector.
Anitated Asier, Gerente de Desarrollo de Negocio EMEA & LATAM en Telefónica Servicios Audiovisuales, y Antonio León Carpio, CEO of Etiqmedia, were responsible for analyzing the role that artificial intelligence plays in areas such as speech technologies (voice recognition, transcription, ASR, detection of audio events, natural language analysis); Image processing (Facial detection, Jewish segmentation, detection of logos and objects); and Natural language processing (Concept analysis, categorization, thesaurus, user profiling, summarization).
In León's opinion, "AI makes sense if it is within the workflows for radio and television, reducing time and providing new values." "An automatic videos processing system is not a black box in which we can put anything and take out anything. We need to segment by type of content and treat them differently and decide which algorithms we apply. In addition, the adaptation for each client is critical," he said.
"It is very important to feedback from the system so that it has continuous training, with new words like Covid, ERTE ... This technology has many limitations. There will never be 100% success but we have to manage the error, correcting it, training the system ... In my opinion they could not put into production algorithms that are below 90% of success rate," he added.
On this webinar, various Use cases in which IA has facilitated processes, increased productivity and cost savings. A typical case we find it on a broader who receive hundreds of Contributions of press conferences daily. With EtiqMedia technology based on algorithms it is possible to segment the video automatically separating interventions, transcribing the different voices, creating labels starting from the radio or television thesaurus or configuring automatic entities based on a semantic analysis.
Another example of automation With etiqmedia, it is a complete informative. With the etiqmedia system it is possible segment All the news so that the Interactive Department can publish the entire informative or by pieces (the system analyzes the semantics, realization, presenters ... to obtain these cuts). The system can integrate the caption Issue from the chain news system (guaranteeing a corrected text), or recognizing it automatically.
It is noteworthy that, according to Antonio León, Etqmedia systems reach a Success rate above 96% of the automatic voice transcription content, being greater if it is regulated content such as a parliamentary session or a press conference. "It is very important that acoustics is correct, since the reverberation and background noise will complicate recognition," he said.
Another aspect analyzed on this webinar was the use of AI in facial recognition, a complex issue for data protection since biometric data is saved.
Etiqmedia has a database of protagonists, and also carries out an internal recognition in the Broadcast. The system is feeded to train the system locally within the BroadCaster network itself. And it is that etiqmedia works on-premise so that everything is inside the client's network, and no content goes out. Only in specific cases in small broadcasters are implanted cloud -based systems.
As for the success in facial recognition, León estimated at 93% thanks to an advanced system that analyzes up to eight versions of a face with different angles or lights. This success rate is being lower due to the generalization of the use of the mask and although the Neuronal Network of Etiqmedia is based on the characteristics of a face (distance between eyes, nose and mouth), based on 128 features, the mask covers part of these facial data hindering the result.
In Broadcast environments, it must be highlighted on the other hand that Etiqmedia is integrated with any average management system that the client has, carrying out the integration in a very simple way.
Institutional scope
Another area in which artificial intelligence gains ground every day is institutional. Etiqmedia solutions already operate in numerous national, autonomous congresses or local corporations. In both face -to -face and telematic plenary, its technology allows you to take out a videoacta, subtitling, or interactive transcription. It is also Multilingual operating in the four official languages of Spain (Spanish, Catalan, Valencian, Galician and Basque) develop. In addition, it develops specific algorithms for each client, which allows, for example, to the Senate to transcribe and catalog their sessions with a 97.8%success rate.
Antonio León said that "transcription automation reduces the time spent five times with respect to an entirely manual process. In fact, the live subtitling has 5 points of success less than the offline. In addition, our technology is capable of subtitulating in bilingual both in Broadcast and institutional environments with a delay of about 3-4 seconds with 90% success."
Highlight, on the other hand, that the impact of Pandemia by COVID-19 has triggered the Online education. Every day there are a bestial amount of videos and there is a need to make them accessible to deaf people or for students in general so that they can recover a class, a conference or specific content among the entire repository.
Other services provided by EtqMedia is the automation of Media monitoring With its Brand Tracking technology. It allows you to know what they talk about us on video, audio, images, text files, social networks ... capturing radio channels, TV and social networks, processing information and obtaining reports about people, matches, companies, places ... of which you are talking about.
Looking to the future
"In Etiqmedia we do not intend to be better than Gooole or Amazon, which are the great world technological, but we do compete strong in niches determined as institutional, informative, and located in markets such as Spain," said the company's CEO.
Finally, he said that the short will implement in Atresmedia A system that will not need the use of metadata since the system will be purely visual. "The images will be passed through the system and it will store visual characteristics of these images and instead of looking for the name, you will seek the image. We will avoid the entire previous labeling process, searching by image," said León.
The next step, he points out will be to look for resources without having to label every time an object comes out for example. Thus, starting from the images already labeled as a car, a terrace of a bar or an airplane, artificial intelligence will be able to look for all similar planes without re -labeled.
In the medium term, Etiqmedia will work in the ABSTRACTIVE SUMMARY OF TEXTS, the Improvement of the Reescale and Video Codification (Improve an image instead of BITS interleave New neuronal networks, complete compression of the scene, the speech enhancement o to Voice and music segmentation coincidents.
Undoubtedly, an entire exciting world that artificial intelligence is opening today and, even more, in the future.
Did you like this article?
Subscribe to our NEWSLETTER and you won't miss a thing.