curveforge

Build Dirac Live target curves from research-grounded recipes.

Shipped target curves overview

curveforge is a small CLI that turns a YAML recipe into a .targetcurve file ready to load into Dirac Live 2/3. It ships with a curated library of in-room target curves (Harman, Olive-Welti, B&K, Toole, Welti-sub, flat) — each parameterised so you can sweep, compare, and tune them rather than being stuck with whatever shelf level a preset shipped with.

pip install curveforge # or: uvx curveforge
pip install 'curveforge[plot]' # adds matplotlib for the `plot` command

Requires Python ≥ 3.11.

A recipe is a YAML file describing a base curve, an ordered list of transforms, and where to write the result.

# my-living-room.yml
output:
 path: living-room.targetcurve
 device_name: Living Room
 low_limit_hz: 10
 high_limit_hz: 24000
base:
 type: harman
 params:
 shelf_level: 8
transforms:
 - peq: { freq: 22, gain_db: 4, q: 1.5 }
 - rolloff: { freq: 15, order: 2 }

curveforge build my-living-room.yml
# wrote living-room.targetcurve

Drop the file into Dirac Live → Custom Target Curve → load.

Worked recipes for common situations. Each one builds cleanly with curveforge build <yaml>; the full versions live in gallery/.

A lean, balanced Harman shelf for accurate mixing-style playback. Light bass weight — the right baseline if you want clarity over physical impact.

Critical listening curve

base:
 type: harman
 params: { shelf_level: 6 }

Harman +10 with the shelf extended into mid-bass (shelf_corner: 150), plus a +4 dB peak at 25 Hz for visceral sub impact. Compensates for the steepening equal-loudness contours at low listening levels.

Cinema curve

base:
 type: harman
 params: { shelf_level: 10, shelf_corner: 150 }
transforms:
 - peq: { freq: 25, gain_db: 4, q: 1.0 }

Modest bass plateau, smooth 0.83 dB/oct treble taper. Anchored without being bloated; smooth without being dark. The closest thing to a "natural room" target.

B&K curve

base:
 type: b_and_k
 params: { bass_level: 3, bass_corner: 200, treble_slope: -0.83 }

Low-shelf with corner at the sub-to-mains crossover. Use this for independent sub-channel correction in Dirac; set Dirac's HIGHLIMITHZ to match the crossover_hz value.

Sub-only curve

base:
 type: welti_sub
 params: { shelf_level: 8, crossover_hz: 80 }

One YAML knob sweeps Harman from lean to cinema — curveforge plot overlays several recipes side-by-side so you can see exactly what each parameter does.

Shelf level comparison

curveforge plot harman-4.yml harman-6.yml harman-8.yml harman-10.yml \
 -o comparison.svg

Want to see how to use these tools to actually voice a system? The Triangle BR09 + SVS SB-2000 case study walks through ~10 listening iterations: each one frames a complaint as a frequency band, diagnoses speaker-voicing-vs-room with the Dirac L+R overlay, applies a targeted recipe change, and reflects on what worked. The point isn't to copy the curve — it's to copy the process.

curveforge list curves and curveforge info <name> print parameters and citations from the CLI. Detailed background on each curve is in the References section at the bottom.

Transforms apply in the order they appear in the recipe.

from curveforge import build_curve, write_targetcurve
curve = build_curve(
 base="harman",
 base_params={"shelf_level": 8, "shelf_corner": 105},
 transforms=[
 ("peq", {"freq": 25, "gain_db": 4, "q": 1.0}),
 ("rolloff", {"freq": 15, "order": 2}),
 ],
)
write_targetcurve(
 path="living-room.targetcurve",
 curve=curve,
 name="My target",
 device_name="Living Room",
 low_limit_hz=10,
 high_limit_hz=24000,
)

Useful for batch jobs, notebooks, or wrapping curveforge in your own tools.

curveforge list curves|transforms|all
curveforge info <curve_or_transform_name>
curveforge validate path/to/file.targetcurve
curveforge diff a.targetcurve b.targetcurve
curveforge render recipe.yml --stdout --format csv
curveforge plot recipe.yml [-o curve.svg] # needs [plot] extra; format inferred from extension
curveforge plot a.yml b.yml -o overlay.svg # overlay multiple recipes
curveforge tweak input.targetcurve tweak.yml # apply transforms to an existing curve

Existing Dirac target-curve resources (the GUI editor, hand-curated target-curve files shared on audio forums) force you to choose between a small set of fixed shapes or hand-edit breakpoints. curveforge gives you the parametric model behind each research target, so you can:

• Sweep a Harman shelf from +4 to +14 dB by changing one number
• Layer a sub-bass peak or a protective rolloff on top of any base curve
• Compare scientific targets at equivalent settings (e.g. Harman vs B&K vs Toole at +6 dB bass)
• Treat target curves as code: version-controlled, diffable, reproducible

A target curve tells Dirac Live what frequency response you want from your speakers in your room. Dirac measures your actual in-room response, then computes correction filters that shape it toward the target. A flat target rarely sounds good — bass loses its weight and the highs feel sharp. The standard recommendations (Harman, Olive-Welti, Brüel & Kjær, Toole) all build in a deliberate bass shelf and a gentle treble tilt because listeners consistently prefer that shape over flat in blind preference tests.

Quick orientation, by listening goal:

• Harman — community-standard, gentle bass shelf around 100 Hz, treble close to flat. Safe default for music; nothing surprising.
• Olive-Welti in-room — slightly more aggressive bass and a steeper treble tilt than Harman. The current Harman research target; often preferred in blind preference studies.
• Brüel & Kjær 1974 — flatter mid/high, more "studio monitor" character. Use when you want a reference, not a flavored sound.
• Toole in-room — gentle downward slope across the whole band, no shelf. Good when your room already loads up the bass on its own.
• Welti subwoofer — sub-only curve; pair with one of the above for the rest of the band.

Try a couple — curveforge makes overlay comparison cheap (curveforge plot a.yml b.yml -o overlay.svg).

REW (Room EQ Wizard) and the Dirac Live app let you edit target curves visually, breakpoint by breakpoint. That works once. If you want to try +0.5 dB on the bass shelf, you redraw the curve.

curveforge inverts the workflow: your target is a recipe — a few parameters or a YAML file. Editing means changing a number. Result: reproducible (lives in git), parametric (sweep and overlay), composable (stack transforms — peaking EQ, shelves, tilts — on any base). REW and the Dirac Live app are still the right tools for measuring and applying correction; curveforge is for designing the target that goes into them.

No. curveforge produces target curves — what you want your room to sound like. Dirac Live (the desktop app) takes the measurements and computes the correction filters to hit that target. You hand Dirac the .targetcurve file curveforge generates. The two tools sit on either side of the measurement step.

Yes. The .targetcurve format is shared between Dirac Live 2 and Dirac Live 3, and curveforge writes the format both versions consume. Tested against Dirac Live 3 on macOS and Windows; should work identically on the embedded Dirac builds shipped with miniDSP, NAD, Storm Audio, and similar hardware.

Each curve module ships with full citations accessible via curveforge info <name>. The condensed sources:

• harman — Olive, S. E. & Welti, T. — multiple AES papers on listener preference for in-room loudspeaker response (Harman International). See also Toole, F. E., Sound Reproduction (3rd ed., 2017), Routledge / AES Presents.
• olive_welti_inroom — Olive, S. E., Welti, T., & McMullin, E. (2013). "Listener Preferences for In-Room Loudspeaker and Headphone Target Responses." AES 135th Convention, paper 8994. (See also paper 8867 at the 134th Convention, which introduces the RR1_G reference curve.)
• b_and_k — Brüel & Kjær Application Note 17-197 (1974), "Relevant loudspeaker tests in studios, in Hi-Fi dealers' demo rooms, in the home, etc., using 1/3 octave, pink-weighted, random noise."
• toole_inroom — Toole, F. E. (2017). Sound Reproduction: The Acoustics and Psychoacoustics of Loudspeakers and Rooms (3rd ed.). Routledge / AES Presents.
• welti_sub — Welti, T. & Devantier, A. (2006). "Low-Frequency Optimization Using Multiple Subwoofers." J. AES 54(5), pp. 347–364. See also Welti, Subwoofers: Optimum Number and Locations (Harman).

MIT