Chant Forge turns Gregorian chant written as GABC text into MIDI, a cue sheet, a printed score and a preview mix. When a chant is too wide for one section of men's voices, it hands phrases between tenors, baritones and basses. Until today the singers then had a score and a cue sheet to learn from, but nothing to put in their ears.
Choirs usually learn parts from learning tracks: one recording per section, with that part louder than the rest. You can buy them per piece from rehearsal-track services, or a director builds them in a DAW, one bounce per part. A split chant needs a set for every section, and they have to be rebuilt whenever the split changes.
What it writes
node render-cli.mjs piece.mid --learn DIR (and a "Download rehearsal tracks" button on the studio page) writes two WAVs per section:
- part: the section loud and panned left, the other sections 12 dB down and panned right, the band 6 dB down. The pan alone puts the singer's line about 6 dB louder in the left ear, so they can follow it without losing the others.
- minus: the whole mix without that section, to sing the part against.
The sections come from the MIDI's own track names, so a tenor/bass split gives four files and a TTBB split gives eight, in score order.
How it was checked
The levels are measured from the rendered audio, not assumed. On a two-phrase test file the part track's own phrase comes out 6.02 dB louder on the left than the right, the other section's phrase is mirrored to the right and sits 9 to 15 dB under it, and the minus track is silent wherever only the left-out section sings. A headless browser downloads the page's ZIP for a split Dies irae and finds all four files byte-for-byte equal to the command line's.
What hasn't been checked: whether singers actually learn faster from these than from the full mix, and whether 12 dB is the right drop. Rehearsal-track makers don't agree on it, which is why --others and --backing set the levels. The "choir" is still the preview synthesiser, so these are for learning notes, not for listening.