High Resolution Guitar Transcription via Domain Adaptation

A new approach was put forward by Maman and Automatic music transcription (AMT) has achieved high accuracy Bermano [9] in which activations from an existing transcription for piano due to the availability of large, high-quality model are used instead of spectral features. In their work datasets such as MAESTRO and MAPS, but comparable this model is bootstrapped using synthetic audio-score pairs datasets are not yet available for other instruments. In recent and then retrained on the aligned scores. This process goes work, however, it has been demonstrated that aligning scores through several iterations of Expectation Maximisation [10] to transcription model activations can produce high quality to improve the results before a final transcription model is AMT training data for instruments other than piano.