Fig. 4From: Classification-based spoken text selection for LVCSR language modelingPerplexities with accumulation of groups for CRF-based scoringBack to article page