Measuring children's phonemic awareness through blending tasks

In this paper, speech recognition techniques are applied to automatically evaluate children’s phonemic awareness through three blending tasks (phoneme blending, onset-rhyme blending and syllable blending). The system first applies disfluency detection to filter out disfluent phenomena such as false-starts, sounding out, self-repair and repetitions, and to localize the target answer. Since most of the children studied are Hispanic, accent detection is applied to detect possible Spanish accent. The accent information is then used to update the pronunciation dictionaries and duration models. For valid words, forced alignment is applied to generate sound segmentations and produce the corresponding HMM log likelihood scores. Normalized spectral likelihoods and duration ratio scores are combined to assess the overall quality of the children’s productions. Results show that the automatic system correlates well with teachers, and requires no human supervision.