Random Utterance Concatenation Based Data Augmentation for Improving Short-video Speech Recognition