January 25, 2018

EEG Breaks Through Automatic Captioning Accuracy Limitations with Development of "Topic Models"

Dave Watts

Now introducing the latest enhancement to EEG's cloud-hosted automatic live captioning platform, Lexi; newly developed "Topic Models" empower the system to recognize topics, immerse itself in distinctive vocabulary, and observe context through the absorption of relevant web data unique to each implementation. This ground-breaking advancement enables Lexi to perform in real-time with a degree of accuracy that reaches beyond previous speech-to-text systems.


As part of Lexi’s setup process, users are now invited to select one of EEG’s developed topic models, or generate their own custom model by supplying Lexi with any combination of reference URL’s or other bulk text data specific to their subject matter or locale. As it absorbs the data, Lexi observes context of new words and names to recognize where and how they are often used. This topic-focused approach ensures that each instance is molded around the subjects, words, names, and phrases intrinsic to the application and provides broadcasters with an unprecedented tailor-fit automated captioning experience; a level of functionality that gives users greater control over situations which routinely produce glaring onscreen errors in other automatic captioning solutions.


“With Lexi’s Topic Models, media customers are spared much of the effort of building and maintaining topic-specific vocabularies; a previously time consuming task.” says Bill McLaughlin, VP of Product Development for EEG. “EEG’s cloud-based libraries provide up-to-the-minute news and sports vocabulary which, when combined with customer-specific talent names, gives unparalleled proper noun recognition.”


Lexi is capable of delivering over 90% accuracy for many common media types - creating a new standard in accessibility for addressing currently uncovered live material in broadcasting, corporate, education, government, and more.