Visual Music Transcription of Clarinet Video Recordings Trained with Audio-Based Labelled Data.