ISCA Archive WOCCI 2008
ISCA Archive WOCCI 2008

Linguistic analysis of spontaneous children speech

Vassiliki Farantouri, Alexandros Potamianos, Shrikanth Narayanan

In this paper, we investigate the duration, lexical and linguistic properties of children's spontaneous speech for children ages 8 to 14 interacting with animated characters in a computer game. Age and gender trends are studied for parameters such as phone and sentence duration, speaking rate, °uency (mispronounciations and hesitations), vocabulary size and linguistic variability measured via bigram language model perplexity. The analysis shows significant differences between read- and spontaneous children speech in terms of absolute values of acoustic and linguistic parameters, as well as, linguistic variability. In addition, spontaneous data present clear gender-specific trends, e.g., increased "language exploration" by girls in the 12-14 age group. The applicability of these results for acoustic and linguistic modeling and spoken dialogue systems interface design is discussed.