On the development of matched and mismatched Italian children's speech recognition systems.

摘要: While at least read speech corpora are available for Italian children’s research, there exist many languages which completely lack corpora. We propose that learning statistical mappings between the adult and child acoustic space using existing adult/children may provide a future direction generating models such data deficient languages. In this work recent advances in development of SONIC recognition system will be described. This work, completing previous one developed past, was conducted with specific goals integrating newly trained into version Colorado Literacy Tutor platform. Specifically, research complete training test set FBK (ex ITC-irst) Children’s Speech Corpus (ChildIt). Using University LVSR system, we demonstrate phonetic error rate 12,0% incorporates Vocal Tract Length Normalization (VTLN), Speaker-Adaptive Trained models, as well unsupervised Structural MAP Linear Regression (SMAPLR).