CITATION — REFERENCE ENTRY
A Robust Mid-Level Representation for Harmonic Content in Music Signals — Proceedings of the 6th International Conference on Music Information Retrieval (ISMIR 2005)
- Key
- bello2005-midlevel-harmony
- Authors
- Bello, Juan Pablo; Pickens, Jeremy
- Issued
- 2005
- Type
- paper-conference
- Container
- Proceedings of the 6th International Conference on Music Information Retrieval (ISMIR 2005)
- Pages
- 304-311
Raw CSL JSON
{
"DOI": "10.5281/zenodo.1417431",
"URL": "https://archives.ismir.net/ismir2005/paper/000038.pdf",
"page": "304-311",
"type": "paper-conference",
"title": "A Robust Mid-Level Representation for Harmonic Content in Music Signals",
"author": [
{
"given": "Juan Pablo",
"family": "Bello"
},
{
"given": "Jeremy",
"family": "Pickens"
}
],
"issued": {
"date-parts": [
[
2005
]
]
},
"event-place": "London, UK",
"container-title": "Proceedings of the 6th International Conference on Music Information Retrieval (ISMIR 2005)"
}
Claims
-
Bello and Pickens averaged the chromagram within beat segments, producing a beat-synchronous representation.
"Finally, the chromagram is averaged within beat segments and further reduced from 36 to 12 bins by simply summing within semitones."
-
Initial HMM transition probabilities favoured triads close to each other on the circle of fifths.
"Triads closer to each other on the circle are more consonant, and thus receive higher initial transition probability mass than triads further away."
-
Bello and Pickens limited their chord lexicon to the 24 major and minor triads.
"we limit our chord lexicon to the 24 major and minor triads"
-
Keeping the musically defined chord observation parameters fixed during training raised accuracy from 42.93% to 75.04% on 28 Beatles recordings.
"The huge accuracy increase (from 42.93% to 75.04%) supports the view that the knowledge about chords encoded in B is universal, and as such it should not be modified during training."
Available in