Using Modified Adult Speech as Data Augmentation for Child Speech Recognition