Section 17 // FA-2026-002 // Forensic Acoustic Pipeline

10-Phase Forensic Pipeline — Multi-Source Acoustic Analysis

Standardized 10-phase methodology applied to the September 10, 2025 UCCU incident. Eleven recordings analyzed across five distinct acoustic sources. Every step documented. Every result reproducible. Methodology framework adopted from established audio forensic practice (Maher 2007 et seq., Knapp & Carter 1976, Cramer 1993).

Case FA-2026-002 // 2025-09-10 // Compiled May 2026 // State of Utah v. Tyler Robinson
01
Intake

Evidence cataloging, hash verification, format inspection. Eleven independent recordings ingested across multiple device classes.

The dataset extends substantially beyond the 4-receiver scope of comparable analyses. Eleven independent recordings have been analyzed in the primary framework, supplemented by the Ken Linke 21-recorder dataset for crack-boom interval cross-validation. Recordings span device classes from consumer cell phones (iPhone 11–16) through professional broadcast equipment (Canon XA55 with 4-channel uncompressed PCM).

IDFileDevice ClassFormatNotable
R1Canon XA55Pro broadcast48kHz 4-ch PCM (uncompressed)733 sub-200µs zero-crossings on CH1 shotgun
R27.mp4iPhone 16 Pro Max48kHz stereo + APAC 4ch spatial123 sub-200µs ZCs (best phone)
R3IMG_6368iPhone 15 Pro44.1kHz stereoWavelet +17dB at 52Hz
R42.MOViPhone 1448kHz stereoCavitation pulses 5.4ms mean interval
R5–R11camera 4, 6, 8, 11, 13, 15, video2_1Various44.1–48kHz stereoMulti-source TDOA cross-validation
FR05miPhone (5m from CK)44.1kHz stereoAudioFreq independent recording (corroborating)
Ken Linke set21 distributed recordersVariousCrack-boom interval database (109–195ms range)
Coverage Comparison

11 primary receivers + 21-recorder cross-validation set versus the 4-receiver scope of comparable opposition analyses. Larger array enables far-field source localization that is geometrically impossible with sub-25m baselines.

02
Recording Characterization

Format analysis, codec inspection, AGC detection, rise-time measurement.

The Canon XA55 recording represents the centerpiece of the dataset. Its uncompressed 4-channel PCM at 48kHz/16-bit preserves the full broadband spectral signature without the lossy AAC artifacts present in cell phone recordings. Where phone codecs compress and decimate high-frequency transient content, the Canon resolves zero-crossing density at the 20.8µs floor (the sample period itself).

Canon XA55 — CH1 Shotgun

733

Sub-200µs zero-crossings (Mach cone N-wave signature). Saturation: 122,844 clipped samples — exclusively on the shotgun channel, indicating on-axis acoustic event directionality.

Best Phone (7.mp4)

123

Sub-200µs ZCs. AAC compression preserves transient timing but smears HF spectral detail.

Phone (IMG_6368)

2

Sub-200µs ZCs. Same recording the opposition relies on heavily. Codec compression substantially degrades fine N-wave structure compared to Canon.

Methodological Implication

Any spectral analysis restricted to phone recordings inherits the codec's HF attenuation. The Canon XA55 PCM is the only path to a defensible broadband spectral measurement. Conclusions drawn from filtered phone audio (e.g., LP500 onset picking) cannot recover information that the codec already removed.

03
Environmental Parameters

Cramer 1993 speed of sound calculation. UCCU site conditions, September 10 2025.

85°FTemperature
37%Humidity
1404mElevation
855.6 hPaPressure
349.7m/s sound speed
# Cramer 1993 — Speed of sound in humid air v_sound = 331.3 × sqrt(1 + T/273.15) × (1 + h_correction) Where: T = 29.44°C (85°F) h = 0.37 (37% RH) P = 855.6 hPa (at 1404m elevation) Result: v_sound = 349.7 m/s # Sensitivity: ∂v/∂T ≈ 0.6 m/s per °C At ±2°F (±1.1°C): Δv = ±0.7 m/s At 100m propagation: Δt = ±0.58 ms At 127m propagation: Δt = ±0.74 ms
Cross-Validation

Environmental parameters and Cramer 1993 result match the TruthEvades FA-2026-001 calculation exactly (v = 349.7 m/s). Same site, same day, same physics. This confirms the underlying environmental framework before any disagreement arises in subsequent phases.

04
Acoustic Event Classification

Multi-source separation. Five distinct acoustic events identified across the dataset.

The 2-event model (crack + blast) used by some opposition analyses is incomplete. The full broadband recording — particularly the Canon XA55 PCM — resolves five distinct acoustic events, each with a characteristic spectral signature, temporal location, and (in three cases) independently localizable origin.

SourceSpectral SignatureTemporal (re: rifle fire)Origin
Mach ConeBroadband (centroid 2,755Hz)t = 0 (reference)Bullet trajectory line
4940Hz StrouhalNarrow-band 4,940Hz+0.5ms after Mach coneVan (3.3m from tent)
Chest Resonance53Hz / 80Hz / 133Hz LF modes+136ms peak (at IMG_6368)Victim thoracic cavity
Stage Detonation656Hz centroid, 104µs rise+187ms at tent / +321ms CanonTent/collar position
Muzzle BlastBroadband (centroid 2,429Hz)+231ms mean (10 cameras)127m rooftop n_wave KML
Why Two-Event Models Fail

A 2-event (crack + blast) classification framework cannot account for: (1) the 4940Hz narrow-band tone localizable to the van, (2) the 53/80/133Hz thoracic resonance peak well before the muzzle blast, (3) the spectrally distinct LF event at +321ms Canon-time. An LP500 lowpass filter applied to phone audio captures the 656Hz detonation and labels it the muzzle blast — but the actual muzzle blast (centroid 2,429Hz) arrives 50ms later and is attenuated by the same filter.

05
Receiver Geolocation

KML placement against satellite imagery. 14 placemarks, ~150m array spread.

Receiver positions extracted from KML survey data with target uncertainty ±0.5m against satellite imagery. Origin: tent center (consistent with KML n_wave reference frame). X = East, Y = North, Z = Up.

ReceiverX (m)Y (m)Z (m)Notes
tent (origin)0.000.000.00Charlie Kirk position
van−2.04−1.400.003.3m from tent — Strouhal source
Canon XA55~+15~+35+1.5~46m from tent — broadcast camera
n_wave (rooftop)+106.94+69.13+21.83Rifle position from KML
11 phonesdistributeddistributed~0Audience ring around tent, ~150m max spread

Array Spread

~150m

Maximum baseline E-W. Sufficient for far-field source localization.

N-S Spread

~120m

Including rooftop n_wave reference at +69m N. Triangulation in 3D.

Geometric Resolution

Larger baseline than 4-receiver opposition arrays (~24m). Direct consequence: this dataset can localize sources at 100+m range; smaller arrays cannot.

Critical Geometric Point

A 24m receiver array cannot localize a source at 127m range to ±2m precision. At far-field distances, all hyperbolas asymptote to bearing lines and degenerate. Tight confidence ellipses on near-field solutions from small arrays are circular consequences of the near-field assumption — not evidence that the source is actually near the array. A 150m array, by contrast, retains hyperbolic intersection geometry at 127m source range.

06
Synchronization

GCC-PHAT cross-correlation with sub-sample parabolic interpolation. Demonstrated ±21µs precision.

# GCC-PHAT (Knapp & Carter 1976) R_12(τ) = IFFT[ X_1(f) · X_2*(f) / |X_1(f) · X_2*(f)| ] # Sub-sample refinement via parabolic interpolation τ_refined = τ_peak + 0.5 × (R[τ-1] - R[τ+1]) / (R[τ-1] - 2·R[τ_peak] + R[τ+1]) # Closure test (for receivers A, B, C) τ_AB + τ_BC + τ_CA should equal 0 # Demonstrated precision in this dataset: # Sample rate (Canon): 48 kHz → 21µs sample period # GCC-PHAT residual: <0.03 samples → <1µs closure # Drift: None detected (±5µs early vs late)
Sync Reference Events

Multiple synchronization references available across the recording set: PA vocal content (matches TruthEvades methodology), the supersonic crack onset (Mach cone arrival as a near-instantaneous reference for collinear receivers), and per-shot optical onset markers from accompanying video frames where present. The Canon XA55 PCM provides the primary timing reference owing to its uncompressed sample timing.

07
Time-of-Arrival Measurement

Manual onset picking with band-isolated filters per source class. Cross-validated against 21-recorder dataset.

Critical methodological choice: onset filters are matched to the spectral character of each event class. Wavelet decomposition (continuous Morlet) for the 53Hz chest resonance. High-pass >1kHz for the broadband Mach cone and muzzle blast. Narrow band-pass at 4940Hz for the Strouhal tone. A single LP500 filter applied uniformly across event classes will misidentify which onset belongs to which event.

EventFilterTOA (re: rifle fire)Cross-validation
Mach ConeHP 1kHz, broadbandt = 0 ref11 cameras consistent within Mach cone geometry
4940Hz StrouhalBP 4800-5100Hz+0.5msTDOA residual 0.000024 (4-camera clean set)
Chest ResonanceCWT Morlet, 30-150Hz+136ms peak+17dB at 52Hz, +14.1dB lung Helmholtz
Stage DetonationHP 200Hz + rise-time+321ms Canon (= +187ms tent)104µs rise time, 656Hz centroid
Muzzle BlastHP 1kHz, broadband+231ms mean (10 cams)Canon 202ms, FR05m 203ms, Ken Linke 109-195ms

4-Phone LP500 Result

  • Crack-blast separation: 10–24 ms across 4 receivers
  • Implied shooter distance: ~13 m (from observer-at-target formula)
  • Implied source position: 2.6 m south of stage center (at the tent)
  • Anomalous vs all other measurements: 5–24× shorter intervals than every other independent dataset

Multi-Filter Result

  • Mach cone-to-detonation: +321 ms at Canon = +187 ms at tent
  • Mach cone-to-muzzle blast: +231 ms (10-camera mean) = 127 m rifle
  • Detonation-to-muzzle blast: +44 ms (separable events)
  • Cross-validated: Maher 240 ms, Canon 202 ms, FR05m 203 ms, Linke 21 recorders 109–195 ms
Why the LP500 Approach Fails

The 656Hz stage detonation centroid sits within the LP500 passband. The 2,429Hz muzzle blast centroid is in the stopband. An LP500 filter preferentially detects the detonation onset and labels it the muzzle blast. The resulting 10–24ms "crack-blast" separation is actually the crack-to-detonation interval, which correctly localizes to the tent — not to the rifle. The rifle's actual muzzle blast arrives 200+ms later, attenuated by the same filter.

08
TDOA Multilateration

Hyperbolic intersection per source. Stationary-source model validity tested via known-source control.

# Stationary source TDOA solver — validity check Test 1: Apply stationary model to BOOM (known stationary source) Input: 11 camera positions + boom arrival times (re: rifle fire) Output: Fitted source = exact n_wave KML coordinates RMS residual = 0.00 ms ✓ → Methodology validated. Boom is correctly identified as stationary. Test 2: Apply identical stationary model to CRACK (Mach cone) Input: 11 camera positions + crack arrival times Output: Fitted source = (-227, 0, -192) m RMS residual = 39.3 ms ✗ → Source 192 meters underground. Physically impossible. → Crack is NOT a stationary source — it is a moving Mach cone.

4940Hz Strouhal TDOA

Source position: (−2.04, −1.40)m from tent
Distance from van: 3.29m
Multilateration residual: 0.000024 (essentially perfect fit)
L/R coherence transition: 0.199 (cam2 9.2m) → 1.000 (cam4 30.4m)
Confirms real point source via near→far-field propagation pattern.

Stage Detonation Timing

Canon-measured: +321ms after Mach cone arrival
Tent-to-Canon travel: 134ms (at 343 m/s, 46m)
Back-calculated origin: +187ms at tent
Bullet arrival at tent: +144ms (127m at 880 m/s)
Detonation occurs 43ms after bullet impact. 50ms before muzzle blast.

Muzzle Blast Cross-Validation

21-recorder Linke set: 109–195ms intervals
Canon XA55: +202ms (357ft equivalent, off-axis)
FR05m mic: +203ms (5m from CK)
10-camera mean: +231ms
All consistent with 127m KML n_wave rifle position.

Stationary-on-Crack Failure

RMS residual: 39.3ms
Fitted source: (−227, 0, −192)m
Result: 192 meters underground
Falsifies stationary-source TDOA on Mach cone arrivals. Methodology that produces this result on the same dataset cannot be trusted for crack-based localization.

09
Uncertainty Quantification

Cross-analyst validation. Independent dataset agreement. Per-source residuals.

Where formal Monte Carlo on individual receiver pairs has been computed (4940Hz Strouhal localization), residuals are essentially perfect. For the broader source-attribution conclusions, cross-validation via independent datasets and independent analysts provides a more demanding test.

SourceMethodResultStatus
Maher (CNN, Sept 2025)Crack-blast TDOA, 1 recording~153m, Losee direction✓ Consistent
Ken Linke 21-recorder setCrack-boom intervals109–195ms range✓ Consistent with 127m
Canon XA55 PCMUncompressed broadband+202ms muzzle blast✓ Consistent with 127m off-axis
AudioFreq FR05mIndependent phase analysis203ms boom, no local overload✓ Corroborates non-local detonation
EpicenterCom multi-source11-camera, 4-method analysis5 sources, 3 locations✓ All TDOA residuals minimized
TruthEvades FA-2026-0014-phone LP500 TDOA2.6m south of stage⚠ Localizes detonation, mislabels as blast
Convergence Across Independent Analyses

Maher (CNN), the Ken Linke 21-recorder set, the Canon XA55 PCM, the AudioFreq FR05m mic, and the multi-source EpicenterCom analysis all converge on a 127m rifle in the Losee direction with crack-to-blast intervals in the 109–240ms range. The TruthEvades 10–24ms outlier is explainable as a methodological artifact of LP500 onset picking on the spectrally distinct stage detonation. Five independent datasets agree. One outlier reduces to a known filter pathology.

10
Reporting

Full methodology documentation. Reproducibility as the standard.

Code Repository

Python pipeline: acoustic_fingerprint_analysis.py
Wavelet decomposition, FFT, beamforming, TDOA solver. Open source.

Raw Data

11 video recordings. KML survey data. Ken Linke 21-recorder spreadsheet. All published or available for independent verification.

Cross-Reference Materials

Canon XA55 4-channel PCM analysis. Section 16 cross-source synthesis. Reasonable doubt presentation. All site sections.

Cross-Source Evidence Synthesis (Section 16)

Five datasets, four methods, comprehensive comparison.

View Section 16

Integrated Forensic Presentation

8-slide deck of all cross-source findings.

Download PPTX

Canon XA55 Analysis

Professional broadcast 4-channel PCM forensic analysis.

View Analysis

// Synthesis

Application of the standardized 10-phase forensic acoustic pipeline to the September 10, 2025 UCCU dataset confirms five spatially distinguishable acoustic sources across eleven independent recordings. Stationary-source TDOA self-validates on the muzzle blast (RMS 0.00ms, exact n_wave KML position) and self-falsifies on the Mach cone (RMS 39.3ms, fitted source 192 meters underground) — confirming that the crack is a moving source that requires retarded-time analysis, not stationary multilateration.

The 4940Hz Strouhal tone localizes to 3.29 meters from the van with TDOA residual 0.000024 and exhibits the L/R phase coherence transition (0.199 → 1.000) characteristic of a real spatially fixed acoustic point source — not an electronic, codec, or resonator artifact. The stage detonation localizes to the tent at +187ms after Mach cone arrival with a 104µs rise time. The muzzle blast localizes to the 127m rooftop position with timing consistent across five independent datasets from three independent analysts.

The opposition's 4-phone LP500 result correctly localizes an acoustic event to the tent at +15ms after the Mach cone — but that event is the stage detonation, not the muzzle blast. Their methodology, applied to a wider dataset with appropriate per-source filtering, recovers the same multi-source framework presented here.