A psychoacoustic-aware IEM correction engine with exact biquad modeling.
Takes stereo L/R measurements and a target curve, outputs optimized parametric EQ presets.
Platform: Python on PC (web interface). Primary output: Poweramp 64-band PEQ (Android).
- 64-band parametric EQ output with exact biquad filter modeling
- Web-based GUI — runs in your browser, no terminal needed
- Auto file detection — drop your measurement files and run
- Auto normalization — finds the flattest region between 500–800 Hz automatically
- Auto freq-end detection — detects usable upper limit from measurement rolloff
- Auto output naming — output file named from your IEM + target automatically
- EQ Bands selector — choose 16 / 32 / 64 bands
- Psychoacoustic optimization — perceptual weighting, Huber robust loss, multi-resolution IRLS
- Exact biquad pipeline — no filter shape approximations
- Post-EQ simulation — always generated; upload
*_PostEQ.txtto Squiglink to verify visually - ZIP output — downloads
result.txt(clean preset),report.txt(full metadata + accuracy), andPostEQ.txtin one file - Built-in target — Euvony Reference (neutral-bright-airy)
Python 3.8+
numpy
scipy
tqdm (optional)
flask
run.batinstalls Flask automatically on first launch. For numpy/scipy/tqdm, run once:
pip install numpy scipy tqdm1. Clone or download this repo
2. Place your measurement files in the Measurement/ folder
Files must be named with [L] / [R] or _L / _R in the filename.
Example: KZ Taurus [L].txt and KZ Taurus [R].txt
3. Double-click gui/run.bat
Browser opens automatically at localhost:5000. That's it.
Left panel — Input:
- Upload your L and R measurement files (
.txtor.csv) - Pick a target curve from the dropdown, or upload a custom one
- Set optional parameters: Freq End, Norm Freq, Norm Mode, Sample Rate, EQ Bands
Click Run
A loading overlay appears with a live progress indicator and elapsed timer.
Right panel — Results:
- Preamp value — displayed prominently in amber. Set this in Poweramp first.
- RMSE — how closely the EQ matches the target
- Download All (ZIP) — downloads 3 files: clean preset, full report, and PostEQ simulation
- Frequency Response chart — Raw L+R avg (grey), Target (blue dashed), Post-EQ (green)
- Accuracy vs Target table — per-octave error, color-coded ✓ / ⚠ / ✗
- Filter Bands — all 64 band values ready to enter into Poweramp
Each run produces 3 files, bundled into a ZIP:
| File | Contents |
|---|---|
IEM - TARGET.txt |
Clean preset — Preamp + Filter lines only, no comments |
IEM - TARGET_report.txt |
Full report — optimizer metadata, RMSE, accuracy table, annotated PEQ |
IEM - TARGET_PostEQ.txt |
Post-EQ simulation — upload to squig.link to verify visually |
Output file format is compatible with Poweramp on Android. Poweramp supports direct import of parametric EQ preset files — no manual band entry needed.
- Download the ZIP and extract
IEM - TARGET.txt - Move the file to your phone storage
- Open Poweramp → three-dot menu → Equalizer → Presets → Import
- Select the
.txtfile — all 64 bands load automatically - Set Preamp to the value shown in the preset — mandatory, do this first
⚠️ Always set Preamp first. Without it, boost filters will cause digital clipping.
All targets are in the targets/ folder.
| File | Character |
|---|---|
Euvony Personal Target.txt |
Neutral-bright-airy. Vocal-forward with extended treble, sub-bass shelf, analytical detail retrieval |
Harman 2019v2 Target.txt |
Consumer-tuned. Elevated bass shelf, warm mids, widely used reference |
IEF Neutral 2020 Target.txt |
Crinacle's original neutral reference. Flat-leaning, less bass than Harman, clinical |
IEF Neutral 2023 Target.txt |
Updated IEF Neutral. Added lower-mid weight, reduced 1–2kHz honkiness, more natural timbre |
#U0394 IEF Preference 2025 Target.txt |
Crinacle's preference curve, not neutrality target. PopAvg-DF (JM-1) base with +10dB bass shelf and -4dB treble shelf. Speaker-like, engaging |
#U0394 5128 DF (Tilt_ -1dB_Oct) Target.txt |
B&K 5128 diffuse field with -1dB/oct tilt. Reference-grade, slightly warm |
#U0394 JM-1 DF (Tilt_ -1dB_Oct) Target.txt |
JM-1 diffuse field with -1dB/oct tilt. Neutral baseline, flatter than IEF Preference |
The default target when no custom target is selected.
Character: Neutral · Bright · Airy
- 20–100 Hz — gentle sub-bass shelf, not bass-boosted
- 150–400 Hz — female chest voice body, elevated for vocal weight
- 500 Hz — slight dip to reduce IEM boxiness
- 1–2 kHz — natural presence, no honk
- 2.5–3 kHz — presence peak, singer's formant region
- 4–10 kHz — maintained for clarity, transient precision, and air
- 10–16 kHz — extended treble rise for airy quality before steep rolloff
Compared to references:
- vs Harman: less bass, brighter presence, much more treble extension
- vs IEF Neutral 2023: more vocal body at 200–400 Hz, higher presence peak, more air above 8kHz
- vs IEF Preference 2025: less bass emphasis, more analytical, brighter overall
# Comments start with #
20.0,96.80
20.3,96.78
...
20000.0,65.12
Supported separators: comma or tab.
AutoEQ CSV format (frequency,raw,smoothed) is also supported — raw column is used.
All input files (L, R, target) must be on the same absolute dB scale.
Recommended — Squiglink:
Upload to squig.link → Normalize → 60 dB → 500 Hz → Download
Optimization stages:
- Gauss-Newton warm start with Jacobian column normalization
- 2-pass IRLS with Huber robust loss and multi-resolution residual decomposition
- Adaptive band reallocation (iterative greedy, up to 4 passes)
- Joint fc+Q+gain optimizer (L-BFGS-B with residual-aware importance sampling)
- Post-joint reallocation check
Regularization: Ridge · Smoothness (1st + 3rd derivative) · Band energy · Q penalty · Perceptual pole radius · HF proximity · Phase slope
Psychoacoustic features: Perceptual weighting · A-weighting masking · ERB-based min-phase blend · Adaptive σ frequency jitter · Multi-resolution IRLS
- squig.link — community database aggregator
- crinacle.com — EARS + 711 measurements
- graph.hangout.audio
MIT — see LICENSE
DAPAAADF (Euvony)
github.com/DAPAAADF