Harmony-Transformer-v2

An improved version of the Harmony Transformer. We evaluated the new model in terms of automatic chord recognition for symbolic music. For more details, please refer to "Attend to Chords: Improving Harmonic Analysis of Symbolic Music Using Transformer-Based Models" (TISMIR 2021).

File Descriptions

BPS_FH_preprocessing.py: preprocessing of the BPS-FH dataset
chord_recognition_models.py: implementations of the three chord recognition models: the Harmony Transformer (HT/HTv2), the Bi-directional Transformer for Chord Recognition (BTC), and the convolutional recurrent neural network (CRNN)
chord_symbol_recognition.py: train the chord recognition models using the 24 maj-min chord representations
functional_harmony_recognition.py: train the chord recognition models using the chord representations of Roman numeral analysis