add OTC model trained on LibriSpeech clean 100 with synthetic transcript errors (error-rate = 0.5)

initial commit