Latin America. Masstech takes advantage of the advantages offered by artificial intelligence for the introduction of metadata in video content. This is one of the company's main offerings, explained Frank Lima, the company's Director of Sales for Latin America.
"The product that is arousing the most interest from the public is MassStore and Disaster Recovery, a metadata implementation enriched through artificial intelligence. Even for customers who already have our solution, it is an exceptional license to access the artificial intelligences of Google, Amazon, IBM Watson and that is why it enriches the metadata automatically," he said during his stay in Mexico.
And he detailed: "Mainly for historical archives, because many of the 50s, 60s, 70s, 80s that in synthesis were digitized, but there is no information, only the name of the file. The artificial intelligence system can automatically listen to the voice and create text. The idea is to identify the places, faces, but in any case recognize the voice, create a text and now that text is metadata that can be searched. For example, if you want to look for someone for information about the president of Mexico in the 70s, maybe there is tape, but you don't know where he is. Just type in the keywords and indicate all the assets where there is a match of that word."
Lima stated that it would be impossible to have a human staff ingesting this type of information for metadata. "To be realistic, companies can't have people to watch thousands and thousands of hours of old videos; it is almost impossible because of time and is very expensive. It's decades of video that you have to watch in real time."
The executive mentioned that Masstech's application has a wide margin of accuracy. "This solution, the English speech recognition part, is on average 90 to 95% effective, and I imagine in Spanish it's more or less the same. It depends a lot on noise, for example; there are recordings that have noise from the tape or that there is noise in the environment. In general, enough information is taken out to have metadata, for example, it could be that the name of the president was mentioned four times and we managed to identify three or one and that is enough," he said.
Text written by Vlad Martínez, special for TvyVideo+Radio.
