Submitted by de Rodrigo 3 The MERIT Dataset: Modelling and Efficiently Rendering Interpretable Transcripts CICLAB Comillas ICAI 10 2