2011 | OriginalPaper | Buchkapitel
Audio Lifelog Search System Using a Topic Model for Reducing Recognition Errors
verfasst von : Taro Tezuka, Akira Maeda
Erschienen in: Database Systems for Advanced Applications
Verlag: Springer Berlin Heidelberg
Aktivieren Sie unsere intelligente Suche, um passende Fachinhalte oder Patente zu finden.
Wählen Sie Textabschnitte aus um mit Künstlicher Intelligenz passenden Patente zu finden. powered by
Markieren Sie Textabschnitte, um KI-gestützt weitere passende Inhalte zu finden. powered by
A system that records daily conversations is one of the most useful types of lifelogs. It is, however, not widely used due to the low precision of speech recognizers when applied to conversations. To solve this problem, we propose a method that uses a topic model to reduce incorrectly recognized words. Specifically, we measure relevancy between a term and the other words in the conversation and remove those that come below the threshold. An audio lifelog search system was implemented using the method. Experiments showed that our method is effective in compensating recognition errors of speech recognizers. We observed increase in both precision and recall. The results indicate that our method has an ability to reduce errors in the index of a lifelog search system.