Tobias Cincarek, Rainer Gruhn
Speech Recognition for
Multiple Non-Native Accent Groups with
Speaker-Group Dependent Acoustic Models
Abstract:In this research, the recognition performance for non-native English speech with different
kinds of acoustic models is investigated. The AMs examined are speaker independent native
models, speaker-dependent models, AMs built from speech data of non-native accent
groups and AMs built from data of speaker clusters. Speaker clusters are derived by
k-means clustering in speaker Eigenspace. Oracle experiments and experiments with
parallel decoding and hypothesis selection are carried out for performance evaluation.