Reading DecayCore Output

In brief

Start with System Health and the summary, then inspect magnitude, filter demand, phase, group delay, and the final impulse. A generated graph is a prediction; confirm the result with a new measurement.

1. What you get after a run

DecayCore produces:

1. Interactive dashboards in the UI (Left/Right analysis view).
2. A ZIP package saved to Documents/DecayCore/filters/<version>/ (safe fallback used if needed):
   • FIR WAV files (either L_...wav + R_...wav or a single Stereo_...wav, depending on layout)
   • Summary report (Summary_<type>_<fs>Hz.txt)
   • Convolver configs (Config_...cfg, camilladsp_...yml)

The ZIP focuses on filters, configurations, and summary data. Use the interactive result graphs in the application for visual inspection.

2. Reading the dashboard (UI)

The result view groups the plots by purpose:

2.1 Magnitude and Alignment

Shows:

• Measured (blue)
• Target (green dashed)
• Predicted (orange)
• Level reference line and smart scan window shading
• Correction range shading (mag_c_min to mag_c_max)

Read it like this:

• Predicted should track target without narrow high-Q spikes.
• Strong LF lift means headroom risk; check Summary headroom section.

2.2 Confidence

Shows measurement confidence on a 0–100% scale. Low confidence reduces correction authority and is a reason to inspect the measurement before increasing correction strength.

2.3 Phase

Shows wrapped phase of the predicted response. Use it mainly to see whether behavior looks smooth around crossover and correction limits.

2.4 Group Delay

Highlights delay structure over frequency.

• Broad LF structures often relate to room modes.
• Sharp local spikes can indicate reflection-related behavior.

2.5 Filter (dB)

This is the FIR gain curve after level-compensated plotting. Use it to confirm boost/cut aggressiveness and low-frequency policy behavior.

2.6 A-FDW Effective BW (oct)

Shows effective adaptive bandwidth when available. If unavailable, the panel displays “No A-FDW BW data (…)”. This is expected in some modes.

3. Reading Summary_...txt

The summary is layered. Main blocks:

3.1 Executive Summary

Includes:

• Acoustic Score (L/R)
• Target Match (percentage + RMS dB)
• Confidence (L/R)
• RT60 wideband (L/R)
• Worst detected event (L/R)

3.2 Core Settings and Analysis Mode

Includes:

• Key settings used for generation
• Analysis mode (native or comparison)
• Comparison grid metadata when comparison mode is active
• Automatic-mode export block when AUTO was used:
  • selected target curve and selection method
  • target-selection trial grid (top-N x trials)
  • cached or searched best preset summary
  • excursion protection seed -> final frequency when auto-tuned
  • winner-polish results for phase_limit and mag_c_min:
    • Phase-limit winner polish: applied (X -> Y Hz, rank A -> B, tested [...] Hz) when the post-search polish improved the winner
    • Phase-limit winner polish: not applied when the original winner was already optimal
    • Mag-c-min winner polish: tested, no change (kept X Hz, tested [...] Hz) when no improvement was found

3.3 Guards and control blocks

Includes:

• Correction Guards (max cut, slope limits, low-bass cut policy)
• Temporal Decay Control (TDC) status and limits
• A-FDW status and effective bandwidth stats
• XO and phase clamp lines

3.4 RT60 and matching blocks

Includes:

• RT60 wideband and optional per-band values
• Confidence and leveling window info
• Target curve match lines for left/right

3.5 Alignment and gain staging

Includes:

• Peak (pre-norm), global offset, auto gain margin, applied auto gain
• Additional sections appended during export:
  • DSP EFFECTIVE PARAMS
  • LEVELING
  • HEADROOM MANAGEMENT
  • BOOST/CUT DIAGNOSTICS
  • CLAMP DIAGNOSTICS
  • STAGE CHECKPOINTS
  • AUTO-ALIGN
  • ACOUSTIC EVENTS (when detected)

4. Practical pass/fail checks

Use these first:

1. Target Match improves clearly versus uncorrected.
2. Net boost peak (post global/headroom) is not positive or only slightly positive.
3. hard_clamp/soft_clip counters are not excessive.
4. No unexpected large tilt in LEVELING.
5. Final Max (filter+auto_gain+headroom) stays sensible for your playback headroom.

5. If the result sounds wrong

• Too little bass:
  • Check low_bass_cut_hz
  • Check effective max boost and blocked boost reason
  • Check correction range upper/lower bounds
• Too bright or sharp:
  • Reduce correction range in upper mids/highs
  • Increase smoothing / reduce aggressive slope behavior
• Unstable imaging or odd timing:
  • Check AUTO-ALIGN delay and gain difference
  • Re-verify measurement consistency

Use re-measurement with filter enabled as the final validation.