Similarities in fundamental frequency in infant speech segmentation models

The present study investigates fundamental frequency as a potential basis for segmentation in models of infant speech segmentation. Pairs of segments that were similar either in terms of fundamental frequency envelop or in terms of transcribed content were found in three different speech styles: speech directed to three-month-olds, speech directed to twelve-montholds and speech directed to adults. Spectral distance between the segments was calculated for each pair and used as a rough measure of spectral similarity. In both the infant-directed speech style conditions, the spectral distance was smaller when fundamental frequency was used as basis for segmentation, compared to when transcriptions were used. In the adult-directed speech style condition, no difference was found between different bases of segmentation.