In this paper, we discussed some existing prob-lems of the method of native listeners' reverse shad-owing and proposed a simple solution so that nativelisteners' reverse shadowing can be used effectivelyas spoken annotations. In the proposed method,in addition to shadowings, readings have to be col-lected from shadowers. However, reading-shadowingDTW experimentally showed to provide scores thatare more highly correlated to perceived shadowabil-ity than GOP scores calculated from shadowings.Further, availability of the DTW-based comparisonis much higher because it can be directly applied tonon-native listeners' shadowings and readings.
展开▼