Confidence Measure for Utterance Verification in Keyword Spotting System

In this article, we propose an utterance verification technique for keyword spotting. The keyword spotting system analyzes a given spoken content and searches every speech segment in which one of pre-defined keywords is uttered. To maintain a stable recognition performance in the system, we propose an utterance verification technique that verifies whether a found utterance, or a candidate keyword segment, can be categorized as a keyword. The proposed approach employs a confidence measure based on the recognition results (N-best log-likelihood). In keyword spotting experiments using spoken broadcast news, our approach achieved superior performance compared to the conventional approach.