RIVIXI
LAB
RIVIXI
LAB
RIVIXI
LAB
RIVIXIAI LAB
Research (UZK)
Rivixi Home / Audio Lab

NDT audio lab

Platform for Verification and Synthesis of WAV Audio Files

An integrated AI & physics ecosystem for authenticity screening of field pipeline WAV recordings, synthetic audio detection, and physically grounded acoustic signal generation for correlators and NDT.

Audio Authenticity Screening & Microstructure AI · 6 Mechanisms

WAV Audio Authenticity Verifier

Multi-mechanism authenticity screening of industrial WAV files across 6 distinct stages: (1) RIFF/WAV structure check, (2) known generator tag identification (RivixiID / LIST INFO), (3) SHA-256 hash & duplicate lookup, (4) spectral & statistical feature analysis (26 parameters), (5) ML ensemble classification (RF, Extra Trees, Gradient Boosting, Isolation Forest), and (6) experimental physical consistency check (inter-channel coherence). Note: generator tags and hash lookups only identify tagged files or exact duplicates; statistical/ML detection is probabilistic and sensitive to post-processing (low-pass filtering, resampling) and unseen generator families.

Format: WAV 16/24/32-bit

Drag & drop WAV file here or click to browse

Supports mono and stereo field records from any equipment (SebaKMT, Vector, Korshun, generators)

Pipeline Inspection Metadata

Parameters for precision calibration, physical data passport, and RIVIXI research archive:

Awaiting Audio File for Inspection

Upload an audio file on the left and click "Verify Authenticity with AI". The system scans 28 spectral features and outputs a screening verdict with cloud routing.

VISUAL DIAGNOSTICS · ACOUSTIC MICROSTRUCTURE

Physical Signal Microstructure: Visual Diagnostics & Authenticity Parameters

Real-time interactive diagnostics of 28 physical microstructure parameters for the uploaded WAV file. Visual inspection of power-line harmonics (50 Hz grid validated; 60 Hz recalibration required), ADC Nyquist filter rolloff, cavitation hydrodynamic turbulence, and waveguide dispersion to screen for known procedural synthesis artifacts.

Physical Feature · 50 Hz Validated (60 Hz Pending)

1. Power-Line Harmonic Contrast (H)

Awaiting WAV analysis...
AWAITING
Measured value:
H_50Hz: —
Upload file to inspect 50 Hz power-line ground loop harmonics (60 Hz US grids require harmonic recalibration).
Criterion: H_50Hz ≥ 3.0 dB (60 Hz recal.) Mechanism: M4
Hardware ADC · Anti-Aliasing

2. Nyquist Spectral Rolloff (ΔS_Nyq)

Awaiting WAV analysis...
AWAITING
Measured value:
ΔS_Nyq: —
Inspection of analog hardware anti-aliasing cutoff filter.
Criterion: ΔS_Nyq ≥ 3.5 dB Mechanism: M4
Hydrodynamics · Turbulence

3. Energy Non-Stationarity (CV_RMS)

Awaiting WAV analysis...
AWAITING
Measured value:
CV_RMS: —
Evaluation of natural hydrodynamic jet cavitation fluctuations.
Range: 0.04 ≤ CV ≤ 0.60 Mechanism: M4
Electronics · Hardware DC Bias

4. DC Offset & LSB Parity (B_DC)

Awaiting WAV analysis...
AWAITING
Measured value:
DC / H_LSB: —
Preamplifier thermal noise and ADC least significant bit toggling.
Parameter: |B_DC| > 0, H_LSB > 0.90 Mechanism: M4
Wave Acoustics · Dispersion

5. Inter-Channel Coherence (γ_xy²)

Awaiting WAV analysis...
AWAITING
Measured value:
MSC: —
Pipeline waveguide acoustic dispersion and soil damping analysis.
Coherence: Welch MSC γ_xy²(f) Mechanism: M6 (Exp.)
AI Ensemble · 4 Models

6. 6-Mechanism Pipeline & ML Ensemble

Awaiting WAV analysis...
AWAITING
Measured value:
RF / ET / GB / Iso: —
Weighted multi-model classifier and anomaly detection output.
Corpus: 1,588 files (LPF-sensitive) Mechanism: M1–M6
Zenodo Preprint (Not Peer-Reviewed)

Evaluated on 1,588 16-bit PCM files (921 field + 667 synthetic across procedural, Physical Leak, and Advanced Physical Leak families). Screening naive procedural fakes only (not standalone anti-spoofing): within-archive 5-fold CV ROC-AUC 1.000; Advanced Physical Leak hold-out recall drops to 0.61 (0.50 cutoff) and 0.00 (0.75); 2nd-order Butterworth LPF at 0.97 f_N drops recall from 0.941 to 0.000; 4/29 (13.8%) external recordings score ≥0.50. Validated on 50 Hz grids (60 Hz recalibration required for US grids).

Read Preprint & Limitations