chorales.doc
- 1. 1. Title of Database: Bach Chorales (time-series). 2. Sources: (a) Chorales: Mainous and Ottman edition. Mainous, Frank D. and Robert W. Ottman, eds. 1966. The 371 Bach Chorales. Holt, Rinehart and Winston, New York. (b) Original owners of database: Darrell Conklin ZymoGenetics Inc. 1201 Eastlake Avenue East Seattle WA, 98102 conklin@zgi.com (c) Donor of database: Same as owner. Ann Blombach of Ohio State University originally supplied me with 4-voice encodings of 100 chorales. The present database is the soprano line, converted into Lisp-readable form, and extensively corrected. 3. Past Usage: Conklin, Darrell and Witten, Ian. 1995. Multiple Viewpoint Systems for Music Prediction. Journal of New Music Research. 24(1):51-73. (Successfully coded chorales in a test set with an average of ~1.8 bits per pitch. Used a learning technique called Multiple Viewpoints.) Abstract: This paper examines the prediction and generation of music using a multiple viewpoint system, a collection of independent views of the musical surface each of which models a specific type of musical phenomena. Both the general style and a particular piece are modeled using dual short-term and long-term theories, and the model is created using machine learning techniques on a corpus of musical examples. The models are used for analysis and prediction, and we conjecture that highly predictive theories will also generate original, acceptable, works. Although the quality of the works generated is hard to quantify objectively, the predictive power of models can be measured by the notion of entropy, or unpredictability. Highly predictive theories will produce low-entropy estimates of a musical language. The methods developed are applied to the Bach chorale melodies. Multiple-viewpoint systems are learned from a sample of 95 chorales, estimates of entropy are produced, and a predictive theory is used to generate new, unseen pieces."
- 2. 4. Dataset synopsis Sequential (time-series) domain. Single-line melodies of 100 Bach chorales (originally 4 voices). The melody line can be studied independently of other voices. The grand challenge is to learn a generative grammar for stylistically valid chorales (see references and discussion in "Multiple Viewpoint Systems for Music Prediction"). 5. Number of Instances: 100 Chorales, each with ~45 events 6. Number of Attributes: 6 (nominal) per event (a) start-time, measured in 16th notes from chorale beginning (time 0) (b) pitch, MIDI number (60 = C4, 61 = C#4, 72 = C5, etc.) (c) duration, measured in 16th notes (d) key signature, number of sharps or flats, positive if key signature has sharps, negative if key signature has flats (e) time signature, in 16th notes per bar (f) fermata, true or false depending on whether event is under a fermata 7. Attribute domains (all integers): (a) {0,1,2,...} (b) {60,...,75} (c) {1,...,16} (d) {-4,...,+4} (e) {12,16} (f) {0,1} 8. Missing Attribute Values: none, repeated sections (|: :|) are not re-encoded, i.e., |:A:|B is encoded as AB. 9. Class Distribution: one class 10. The grammar describing the chorale dataset: Dataset -> Chorale* Chorale -> (Chorale_Number (Events)) Events -> Event Events Events -> Event Event -> (Attributes) Attributes -> (st S) (pitch N) (dur D) (keysig K) (timesig T) (fermata F) (see 7. above for attribute domains)