fb

**fb is Humdrum's figured-bass tool run once per chorale: it reduces the harmony at each moment to the intervals every other voice forms above the bass, the way a figured-bass symbol would.

fb is one spine per chorale, not one per voice

Unlike other features, fb isn't a property of a single voice, it's a property of the whole texture at a moment in time. So there's only one **fb spine per chorale, aligned with the bass line, and every lookup against fb resolves to it regardless of which voices you asked for in voices. Searching fb with "voices": "soprano" still reads the same chord-above-the-bass data as "voices": "bass". voices only affects which voice's kern/deg/mint onsets you're stepping through if fb is a secondary constraint rather than the driving feature.

How the **fb spine is generated

The chorale-search binary runs humlib's figured-bass tool once per chorale with the flags -c -n --hint:

• -c (compound) folds compound intervals down to within one octave, so a 10th above the bass is reported as a 3rd, an 11th as a 4th, and so on. There's no way to distinguish a 10th from a 3rd in the resulting data.
• -n (normalize) removes the figures 1 (unison) and 8 (octave), removes duplicate figures (the same interval sounding in two upper voices at once collapses to a single figure), and sorts the remaining figures by size. The sort order doesn't matter for searching.
• --hint is what makes each figure carry an interval-quality letter (M, m, P, A, d, …) instead of a plain figured-bass numeral. It's why a token looks like m6 m3 rather than a bare 6 3, and is what makes the figure- and quality-optional matching below possible.

Special case: 9. A compound 2nd is only ever reported as 9 when the same chord also contains a (compound-reduced) 3; otherwise it stays 2, even though both are the same interval folded down by -c. If you want to be sure not to miss a suspended 9th regardless of whether the chord happens to also carry a 3rd, search for both figures:

{ "fb": ["7 9", "2 7"] }

Token format

A **fb token is a chord: one or more space-separated intervals, each an optional quality letter (M, m, P, A, d, …) followed by a diatonic figure number, e.g. m6 m3 or P5 M3 (a root-position major triad). The figures don't say which upper voice sings which interval. fb only records which intervals are sounding above the bass, not their distribution across alto/tenor/soprano.

Figure- and quality-optional matching

Just like mint, an fb pattern value doesn't have to fully specify a figure. The quality letter can be omitted to match any quality:

And a pattern value can itself list several space-separated figures, e.g. "M6 3" or, order doesn't matter, "3 M6". Both mean "the chord contains a major 6th and a (any-quality) 3rd". By default this is a minimum requirement: a pattern of "2 4" also matches an actual chord of "2 4 6", since both 2 and 4 are present. Use fbCompareExactChord if you want to require the chord to contain exactly those figures and no others.

{
    "feature": "fb",
    "pattern": [
        { "fb": "A6 M3" }
    ]
}

finds every augmented sixth chord in the corpus, regardless of which voice carries which figure.

Using it in a pattern

fb is most useful either as the driving feature (walking chord-to-chord through the harmony) or as a secondary constraint alongside kern/deg/mint (checking "what's the harmony doing when this melodic thing happens"):

{
    "feature": "deg",
    "voices": "soprano",
    "pattern": [
        { "deg": "7", "fb": "6" },
        { "deg": "1", "fb": ["5 3", "3"] }
    ]
}

matches a soprano leading-tone-to-tonic resolution (7 → 1) where the leading tone is harmonized with a 6th above the bass and the tonic lands on a root-position triad (or bare 3rd). See Patterns for OR-lists, wildcards, multi-feature positions, and negation, all of which work the same way for fb as for any other feature.