KZ EDX Pro
- Folder
- iem/
- Release
- HD core
- IRS
- 4-ch, 2851 taps, 48 kHz
- DDC
- 14 biquad sections
- Driver
- 24 Ω, 112 dB/mW
- EEL
- Lookahead limiter
The IEM voicing is a labelled design choice, not a measurement-derived target.
ROOTLESSJAMESDSP · 48 KHZ · TWO DEVICE PROFILES
Purpose-built DSP chains for the KZ EDX Pro and Redmi 12 speaker. Download the core trio or the full library, verify files in your browser and compare chains live.
Designed & tuned by Noushad M. · Offline file verification · Device testing status disclosed
HyperAudio publishes two device profiles, and their deployment is not identical. Both core profiles run the three-file chain below. The KZ EDX Pro HD core uses a 4-channel true-stereo IRS with a tonal DDC. The Redmi 12 V6 core uses an identity DDC with a stereo IRS. The standalone PEQ setup for the Redmi 12 lives in the library.
One correction stage only. On the IEM the DDC and IRS are tuned as a pair, and on the Speaker all correction lives in the IRS. Never stack library files on a core set.
Speaker split. The standalone Liveprog PEQ in the library replaces the VDC and IRS chain, so do not load the VDC or IRS with it.
The IEM voicing is a labelled design choice, not a measurement-derived target.
The V6 three-file core is the default export. The standalone PEQ setup described in the guide is kept in the library.
speaker/core/ ├── SPK_v6.vdc ├── SPK_v6.irs └── SPK_v6.eel
iem/core/ ├── EDX_HD.vdc ├── EDX_HD.irs └── EDX_HD.eel
Hz (log) · dB absolute · 1/6-octave smoothed · measured from the hosted IRS
The full profile library for both devices. Sets are ordered newest first, and each set lists its DDC, IRS and EEL files. Pair files only within one set. The core downloads above are unchanged.
The hosted files are fetched and checked in your browser: IRS format, finite samples, symmetry, peak ≤ 0 dBFS, max gain < 0 dB, L1 ≤ 1.25, tail ≤ −80 dBFS; DDC stability; EEL structure and slider ranges.
Offline checks on both chains. Not app, ARM or device validation.
| Component | Offline / numerical | App or device | Listening |
|---|---|---|---|
| Speaker VDC | PASS identity | USER REPORTED | USER REPORTED |
| Speaker IRS | PASS | USER REPORTED | USER REPORTED |
| Speaker EEL | PARTIAL compile, ceiling, bypass; non-finite rerun pending | USER REPORTED | USER REPORTED |
| IEM VDC | PASS stable | — | — |
| IEM IRS | PASS | NOT RUN | NOT RUN |
| IEM EEL effects path | Prior v4 checks: effects, ceiling, bypass, non-finite | NOT RUN for current v5 | NOT RUN |
| IEM EEL v5 mode selector | Source review only; VM retest pending | NOT RUN | NOT RUN |
Evidence boundary. Offline checks use a source-built JamesDSP VM on x86_64. They are not native ARM, RootlessJamesDSP app, device or listening tests. Numerically qualified does not mean device validated.
Each correction is fitted against a target curve, in dB relative to 1 kHz. Both targets are custom, built offline, and not published standards. Listening results outrank the curves.


Method and principles: Raw and target curves.
addons/ holds two optional impulse responses and three EasyEffects reference presets. See the add-on guide.
BBE SonicMax.irs +7.9 dB (left) and +9.8 dB (right); EFOtech MLV 01024 (True Stereo, 48kHz).irs +3.7 to +5.3 dB across its four channels. The project gate is 0 dB. The convolver loads one IRS at a time, so an add-on IRS replaces the tonal-correction IRS rather than adding to it. Lower the volume before trying one.
Optional impulse responses to replace the core IRS with hardware-accurate phase alignment or spatial crossfeed. Use instead of the core IRS, never alongside it.
Reference files for Night Listening, Concert Hall, and Gentle Dynamics. These cannot be loaded directly into RootlessJamesDSP; they are for manual parameter replication.
Chain: DDC → Convolver (IRS) → EEL2 → Output. Set the host sample rate to 48 kHz and load each file into its slot.
Tonal correction lives only in the IRS. Never load the same EQ in both the VDC and the convolver: it applies twice.
Checks ran offline (source-built JamesDSP VM on x86_64 and NumPy). Not yet run: the app, native convolver, ARM, null tests and listening tests. The IEM tuning has no measurement behind it and the speaker EEL NaN/inf test is pending a rerun.
The browser checks on this page are file checks, not RootlessJamesDSP runtime validation.
Read the repository documents here. Each one also links to its file on GitHub.
Verify downloads with sha256sum. Tap a row to copy its hash. Library hashes are listed in SHA256SUMS and on each card.
Presets derive from AutoEq-format data (jaakkopasanen/AutoEq, MIT) via RikudouGoku's database, corroborated by Speaker Ranking. EEL2 scripts target the VM in RootlessJamesDSP (ThePBone, Apache 2.0). No third-party code or raw measurements are redistributed. See ATTRIBUTION.md.
HyperAudio/ ├── README.md · index.html · ATTRIBUTION.md · LICENSE* ├── iem/ core/, library/, guide, report ├── speaker/ core/, library/, guide, report ├── data/ assets.json, assets.meta.json, curves ├── tools/ build_assets.py ├── docs/ vdc_functions, irs_functions, eel_functions, │ system_full, methodology, test_report ├── assets/ logo, banner, style.css, script.js, docs.js, enhance.js, │ library.js, compare.js, │ fonts, verification harness graphics └── .github/workflows/ci.yml