作者: D. Giuliani , M. Gerosa
DOI: 10.1109/ICASSP.2003.1202313
关键词:
摘要: Recognition of children's speech was investigated by considering a phone recognition task. Two baseline systems were trained, one for children and adults, exploiting two Italian databases. Under matching conditions, training performed with data from the same population group, accuracy 77.30% 79.43% respectively. It found that, many children, results as good adults. However, higher variability in across speakers observed than Vocal tract length normalization, under matched mismatched testing also investigated. For both adults performance improvement, respect to systems, observed.