Figure 10From: Phoneme and Sentence-Level Ensembles for Speech RecognitionWord error rates for various values of , compared with the phoneme boosting approach, for the training and the holdout set. TrainingHoldoutBack to article page