Enriched technology-enabled annotation and analyses of child speech

This paper reviews a range of studies illustrating the ways in which speech technology has enabled richer analyses of corpora of young children’s speech and infants’ speech-like vocalizations. One set of studies illustrates the use of speech synthesis technology, such as VLAM, an articulatory synthesis system that models the transfer functions of childor infantproportioned vocal tracts. VLAM has been used to evaluate cross-language differences in the emergence of contrasting vowel categories. Another set of studies illustrates the use of modern corpus development and annotation tools to create and analyze the paidologos corpus, a databse of utterances elicited in a picture-prompted word-repetition task from 2through 5year-old child speakers of a variety of languages. Flexible, incremental annotation on multiple tiers allows researchers to extract target sounds for spectral analysis as well as for calculating transcribed accuracy rates. Tag sets also can be used to extract stimuli for perception experiments, yielding naive-listener responses that can become another layer of tags.