Multi-Modal Learning for Speech Emotion Recognition: An Analysis and Comparison of ASR Outputs with Ground Truth Transcription.