Accuracy of the Language Environment Analysis (LENA) speech processing system in identifying communicative vocalizations of young children and adults

Abstract

The Language Environment Analysis (LENA) system is an automated audio processing system widely used for characterizing language behaviors of children and adults for clinical and basic research. While a number of studies have assessed LENA’s reliability, its accuracy at identifying and counting speech communicative events is still not well-characterized under a range of naturalistic conditions. In two studies, we examined accuracy of LENA's speech vocalization classifications, relative to human gold standard coding for audio events, as well as word and speech vocalization counts for adults and child utterances, respectively. We found that the weighted average of accurate classification of 100-msec frames by LENA for child speech, adult female speech, and adult male speech was 57%, 61%, and 57%, respectively. Further, an analysis of LENA’s ability to accurately discriminate frames of speech vocalizations from a “key child”—a child wearing the LENA device—from speech vocalizations of other child and adult talkers and sound sources showed that LENA correctly detected key child speech vocalization frames only 41% of the time. We are currently extending this research to examine the accuracy of LENA’s child vocalization count (CVC) and conversational turn count (CTC) measures.