A study of irrelevant variability normalization based training and unsupervised online adaptation for LVCSR.

摘要: This paper presents an experimental study of a maximum likelihood (ML) approach to irrelevant variability normalization (IVN) based training and unsupervised online adaptation for large vocabulary continuous speech recognition. A moving window frame labeling method is used acoustic sniffing. The IVN-based achieves 10% relative word error rate reduction over ML-trained baseline system on Switchboard-1 conversational telephone transcription task.