[ArianMoeini]

Closes #13.

Summary

Add a runnable, GUI-free walkthrough that creates seeded stereo noise and a deliberately filtered reference, then uses the existing engine to analyse both signals, design a matching FIR, render the result, and compare spectral shape. The example can optionally write generated 32-bit float WAVs to the already ignored outputs/ directory.

The example README explains power versus amplitude dB ratios, octave smoothing, match amount, gain limits, and requested versus actual FIR response. The root README links to it.

Observed result

With seed 42, 8 seconds at 48 kHz, 100% amount, 0.33-octave smoothing, ±18 dB limits, and 513 taps:

• Spectral-shape error (100 Hz–15 kHz, log-frequency RMS with overall level removed): 5.49 dB before, 0.06 dB after, a 98.8% reduction.
• Requested versus actual FIR response: 0.03 dB RMS over the same band.
• Optional source, reference, and processed WAVs were verified as 48 kHz, 384,000-frame, two-channel, 32-bit float files.

Verification

• ruff check . passed.
• 69 non-GUI tests passed, including all six engine tests.
• The example ran with and without --output-dir; generated media remains ignored by Git.
• The full pytest command could not collect the GUI suites in this checkout because PySide6 and pyqtgraph are not installed. No GUI code was changed.