Jazz Harmony Treebank

This repository contains the Jazz Harmony Treebank, a corpus of hierarchical harmonic analyses of jazz chord sequences selected from the iRealPro corpus published on zenodo by Shanahan et al.

Attribution

If you use this data in any way, please cite the the following paper:

D. Harasim, C. Finkensiep, P. Ericson, T. J. O'Donnell, and M. Rohrmeier (2020). The Jazz Harmony Treebank. In Proceedings of the 21th International Society for Music Information Retrieval Conference, pp. 207-215. Montréal, Canada. https://doi.org/10.5281/zenodo.4245406

Data format description

The treebank is available in the file treebank.json, structured as a JSON with the following fields:

• title: The title of the tune
• chords: The list of chord symbols. The length of this list is the same as the lists of measures and beats. Chord symbols are strings composed of
  • A root note (an uppercase letter from A to G with optional accidential # or b.
  • A basic chord form. For minor chords a lowercase m, for augmented chords a +, for half and fully diminished chords a % and o, respectively, and for suspended chords a sus. Major chords are denoted without a symbols for its basic chord form.
  • An optional 6, 7 or ^7 indicating an added sixth, minor, or major seventh, respectively.
• measures: A monotonically increasing list of integers that indicates the measures of the tune. Measures which appear more than once indicate that they comprise more than one chord symbol.
• beats: A list of beats with chord onsets. There is always a 1 item for each measure, denoting the chord played on the measure's downbeat. Additional entries are integers that indicate the beats on which subsequent chords within the same measure start. For example in common time, a measure with two chords on the beats 1 and 3 corresponds to the beats list [1,3].
• turnaround: An integer that indicates which chord in the sequence is the final tonic chord of the tune. If the number is 0, then it is the final chord of the chords list. A positive value of 2 indicates that the third last chord is the tonic, while the last two chords constitute a turnaround to bring the tune back to the beginning for repeated choruses. A negative number of -1 indicates that the tune should be repeated and end on the first chord from the beginning of the chords list to end with an appropriate tonic.
• trees : A list of all the tree analyses for a tune. If this element exists, it contains at least one tree analysis. Each analysis is a JSON object consisting of of two elements: open_constituent_tree and complete_constituent_tree. Both contain harmonic tree analyses of the full harmonic sequence, as ended at the chord indicated by the turnaround annotation. The difference between the two trees is further described in the correponding paper (see above). In short, the open_constituent_tree contains phrase analyses additional to the harmonic reference of the complete_constituent_tree. Both trees are JSON objects recursively defined as having a label (which is a chord taken from the chords list), and a list children of either zero or two subtrees, which are themselves trees.
• comments: A string containing comments from the annotation procedure, for example noting alternate chord sequences from other sources, fixed transcription errors, ambiguities, chord roles, etc.
• composers: A string with the name(s) of the composer(s) of the tune, as recorded in the iRealPro corpus.
• year: An integer representation of the year (CE) in which the tune was composed according to the iRealPro corpus.
• meter: A JSON object containing two integers, numerator and denominator indicating the meter of the tune.
• key: The key of the tune, as annotated in the iRealPro corpus. This is an uppercase (for major keys) or lowercase (for minor keys) letter between a-g, possibly followed by a - to indicate a lowering accidental.

Utility library

We include a basic set of utility functions for interacting with the dataset in jht_utilities.py. In particular, it contains the following functions:

• qtree_to_dict: Convert a tree in qtree string form to a Python dict in the format of the treebank. For example, convert [.A B [.C D E]] into {'label': 'A', 'children': [{'label': 'B', 'children': []}, {'label': 'C', 'children': [{'label': 'D', 'children': []}, {'label': 'E', 'children': []}]}]}.

• dict_to_qtree: Convert a tree in treebank format into a qtree string.

• plot_qtree and plot_dict: Use pdflatex and tikz to generate a pleasant visualisations of a single tree. In addition to the tree, this function takes also a filename to which the tree should be written, and optionally whether or not some basic escaping of LaTeX symbols should be applied to the node labels.

• For the following, we assume the tree to be in dict form, as given for example by loading the treebank using json.load.

• contains_open_constituents: Traverse the tree to look for nodes marked as open constituents (see the paper cited above for details).

• unfold_open_constituents Traverse the tree from the root to the leafs and transform it to obtain the pure harmonic reference structure. This function removes all open constituents.

• leaf_labels Obtain the trees leaf labels as a string. For instance [B, D, E] in the above example.

Tree figures

Have a look at the file tree-plots.md to see the plots of all tree analyses.

Dataset statistics

Below are general summary statistics about the dataset as described in the corresponding paper:

The first plot shows that the analyzed pieces is chosen relatively independently from the year of composition. The second plot shows the bias for short pieces in this subset. The [fourth] (the third plot was omitted from the paper) plot shows that the length of turnarounds, if present, usually ranges between 1 and 3.

The two last plots show separately for major and minor keys how often a context-free grammar rule is used in the hierarchical analyses. For these plots, all chord sequences were transposed to C major or to C minor, respectively. Prolongations of the tonic, preparations of the tonic by the fifth scale degree, and preparations of the fifth scale degree by the second are by far the most common rules.

References

The iRealPro dataset that this research builds on was created by the user community of the iRealPro app and first presented scientifically in

Daniel Shanahan, Yuri Broze, and Richard Rodgers (2012). A Diachronic Analysis of Harmonic Schemata in Jazz. In Proceedings of the 12th International Conference on Music Perception and Cognition and the 8th Triennial Conference of the European Society for the Cognitive Sciences of Music, pages 909–917.

Acknowledgements

This project has received funding from the European Research Council (ERC) under the European Union's Horizon 2020 research and innovation program under grant agreement No 760081 – PMSB. We gratefully acknowledge the support of the Natural Sciences and Engineering Research Council of Canada (NSERC), the Fonds de Recherche du Québec, Société et Culture (FRQSC), and the Canada CIFAR AI Chairs program. We thank Claude Latour for supporting this research through the Latour Chair in Digital Musicology. The authors additionally thank the anonymous referees for their valuable comments and the members of the Digital and Cognitive Musicology Lab (DCML) for fruitful discussions.