I. Psychology of False Perception

Why we hear things that aren’t there

Top-Down Processing

The brain doesn’t just listen—it constructs perception

• When audio is unclear, the brain fills in gaps
• External suggestions become “heard” speech
• This process is subconscious and involuntary
• Even experts cannot resist it through willpower

Textual Priming

A transcript doesn’t assist perception—it creates perception

Once a listener sees a word on a page while listening to indistinct audio, they often hear that word “with their own ears”—even if the transcript is demonstrably wrong.

The “Crisis Call” Study

Contextual Priming

Background knowledge about a case biases what we hear

• Knowing someone has a criminal history
• Believing a confession exists
• Having a theory of guilt

→ Listeners hear incriminating words in innocuous sounds

Confirmation Bias

People interpret ambiguous audio to support existing hypotheses

• Evidence against the theory is discounted
• Contradictory acoustic evidence is ignored
• Initial interpretations become entrenched
• “Corroboration inflation” spreads bias to other evidence

Fluency Misattribution

The ease of processing is mistaken for proof of accuracy

1. Transcript makes unclear words easy to “hear”
2. Brain experiences processing fluency
3. Fluency feels like recognition
4. Listener concludes: “That must be correct”

The “Ad Hoc Expert” Problem

Courts allow police to provide transcripts because they’ve listened “many times”

The assumption: Repetitive listening confers expertise

The reality:

• Repetition entrains initial (often wrong) perceptions
• Investigators’ case knowledge primes their hearing
• Their “ability” to hear more is actually bias

II. Case Studies

Real lives destroyed by flawed audio evidence

David Eastman (Australia)

• Convicted: 1995 for the 1989 murder of a police commissioner
• Audio evidence: Thousands of hours of covert recordings from hidden devices in his home—whispered self-talk
• The transcript: Police claimed he said "I killed Winchester"

David Eastman: What Went Wrong

• Police acted as “ad hoc experts” on their own recordings
• Audio was extremely poor quality, whispered
• A phonetic expert later found the audio was consistent with: “I kept watching her”
• Trial judge endorsed police transcripts after “checking them personally”

David Eastman: Outcome

• 19 years in prison
• Conviction quashed in 2014
• Found not guilty at 2018 retrial (misleading transcripts excluded)
• Awarded $7+ million in compensation

Man wrongfully jailed for 20 years wins $7 million compensation | ABC News

The “Pact” Case (R v. Clark)

• Convicted: Father as "accessory before the fact" for murder committed by his son
• Audio evidence: Whispered covert recording from a hidden device in the family home
• The transcript: "At the start we made a pact"

The “Pact” Case: What Went Wrong

Phonetic analysis revealed:

• Rhythmic structure consistent with “it’s fucking payback”
• The word “pact” was never spoken
• Jury, primed by transcript, believed they heard it “with their own ears”

The “Pact” Case: Outcome

• Father sentenced to 30 years—10 years longer than his son
• Conviction stood despite proven error
• Reason: All legal procedures had been followed correctly

Guy Paul Morin (Canada)

• Convicted: 1992 for the 1984 murder of his 9-year-old neighbor
• Audio evidence: "Earwitness" identification—victim’s mother claimed she heard Morin’s voice crying "Help me, help me, oh God, help me!"

Guy Paul Morin: What Went Wrong

• Mother was only a casual acquaintance of Morin
• Commission later deemed the identification “patently unreliable”
• Voice identification by non-familiars is notoriously inaccurate
• Exonerated by DNA in 1995

David Shawn Pope (Texas)

• Convicted: 1986 for aggravated sexual assault
• Audio evidence: "Voiceprint" spectrogram analysis comparing phone calls from the rapist to Pope’s voice
• The analyst: A police officer with two weeks of training

David Shawn Pope: What Went Wrong

• “Voiceprint” methodology later discredited by National Science Foundation
• No scientific basis for spectrogram matching
• 15 years in prison
• Exonerated by DNA in 2001

Russell Faria (Missouri)

• Convicted: 2013 for murdering his wife
• Audio evidence: 911 call "analysis" claiming to identify guilt based on word choice and tone
• The claim: Faria was too "me-focused" on the call

Russell Faria: What Went Wrong

• “911 call analysis” is junk science with no empirical validation
• Used to override strong alibi and lack of physical evidence
• Acquitted at 2015 retrial after:
  • Real killer identified
  • Hidden police photos discovered that contradicted prosecution theory

Michael Williams (Chicago)

• Jailed: One year for murder he didn’t commit
• Audio evidence: ShotSpotter gunshot detection alert
• The problem: "Black box" algorithm with minimal human oversight

Michael Williams: What Went Wrong

• ShotSpotter frequently mistakes car backfires and fireworks for gunfire
• System provided two different locations over a mile apart
• Human reviewers have minimal training
• Charges dismissed in 2021 due to insufficient evidence

Litigating Shotspotter Evidence

Chicago police no longer alerted by ShotSpotter gunshot detection technology starting at midnight

Other ShotSpotter Commentary

• listening in on the neighborhood - Article discussing privacy and reliability concerns with ShotSpotter
• shotspotter-locations - GitHub repository mapping ShotSpotter sensor locations
• Source Localization with ShotSpotter Audio - ShotSpotter’s research on audio source localization

Stefan Kiszko (UK)

• Convicted: 1976 for the 1975 murder of 11-year-old Lesley Molseed
• Audio evidence: Coerced confession after three days of interrogation without legal representation
• Key fact: Kiszko had a mental age of 12

Stefan Kiszko: What Went Wrong

• Investigators had tunnel vision—focused on social awkwardness
• Ignored medical evidence: Kiszko had hypogonadism
• This made it medically impossible for him to produce the sperm found at the crime scene
• 16 years in prison; conviction quashed 1992
• Died of heart attack 22 months after release

Common Patterns

III. Voice Identification Problems

The science says it’s far less reliable than we assume

Voice vs. Face Recognition

Under similar conditions, voice recognition is consistently less accurate than face recognition

• Voice is a weaker memory trace
• The brain prioritizes visual information
• Confidence does not correlate with accuracy

Familiar vs. Unfamiliar Voices

The Face Overshadowing Effect

Seeing a face impairs voice recognition

• Brain prioritizes visual over auditory identity cues
• Effect persists even when witnesses are told to focus on voice
• The bias is not under conscious control

Cross-Linguistic Challenges

Unfamiliar accents and languages dramatically reduce accuracy

• “Other-accent” effect: ~20% accuracy drop for different accents
• Foreign languages: Accuracy falls to 45%–60%
• These rates are insufficient for forensic certainty

Implications for Courts

• Earwitness testimony should be treated with extreme caution
• Unfamiliar voice identification is barely better than chance
• Confidence is not an indicator of accuracy
• Juries must be warned about these limitations

IV. Prevention Methods

Protocols and practices that can prevent wrongful convictions

Core Principle

Treat forensic audio as scientific discipline, not “common knowledge”

• Formal methods, not intuition
• Independent analysis, not investigator transcripts
• Documented procedures, not ad hoc judgments
• Acknowledged uncertainty, not false confidence

Linear Sequential Unmasking (LSU)

Analyze evidence before learning case context

Protocol:

1. Examine raw audio in complete isolation from case information
2. Document initial findings
3. Only then reveal context in controlled stages
4. Separate each stage of analysis

LSU in Practice

Separation of Roles

Forensic analysis must be independent from investigation

• Analysts should not be employed by police
• Analysts should not know the “desired” outcome
• Accredited practitioners only
• Double-blind procedures for voice lineups

Transcription Protocols

End the practice of police-produced transcripts

Requirements:

• Independent transcribers employed in public service
• Isolation from specific investigation
• "[Inaudible]" for unclear words—no guessing
• Acoustic-phonetic verification of auditory findings

Expert Requirements

Scientific recognition: Transcription is linguistic science, not legal precedent

Epistemic modesty: Acknowledge limits and potential for error

Core competencies:

• Track auditory acuity regularly
• Deep knowledge of EMI mitigation
• Signal path design expertise
• No statements of “absolute certainty”

Jury Instructions

Current instructions often fail—timing and language matter

Effective practices:

• Present instructions before evidence is played
• Use simple, explicit language
• State clearly: “Transcripts are not evidence”
• Warn that confidence ≠ accuracy
• Note unfamiliar accent/language unreliability

Technical Standards

Audio format: Uncompressed PCM (WAV), 16-bit minimum, ≥16 kHz sampling

Hash verification: MD5 or SHA to confirm data integrity

Chain of custody: Document every transfer and access

Enhancement caution: “Clearer” audio is not necessarily more intelligible—can boost false transcript credibility

ENF Analysis

Electric Network Frequency: The power grid “hum” embedded in recordings

Uses:

• Verify date and time of recording
• Detect butt-splices and edits
• Identify broad geographic location (different grids have different patterns)

ENF Limitations

• Recording must have been near power grid or electromagnetic field
• Difficult with low signal-to-noise ratio
• Heavily compressed audio may not work
• Requires access to specialized reference databases
• Not a silver bullet—one tool among many

V. Recommendations

What practitioners, courts, and experts should do

For Forensic Practitioners

1. Use LSU for every analysis—context last, not first
2. Document everything—settings, tools, versions, steps
3. Mark unclear segments as inaudible—never guess
4. Verify auditory findings with acoustic analysis
5. Acknowledge uncertainty in all reports
6. Track your own accuracy over time

For Courts and Attorneys

1. Exclude police-produced transcripts as unreliable
2. Require independent forensic analysis
3. Demand documentation of methods and uncertainty
4. Instruct juries before playing audio, not after
5. Warn juries about transcript priming effects
6. Scrutinize “novel” audio analysis methods

For Expert Witnesses

1. Never claim certainty you don’t have
2. Explain limitations clearly to juries
3. Resist pressure to strengthen conclusions
4. Distinguish observations from interpretations
5. Be prepared to say “I don’t know”
6. Prioritize truth over advocacy

Red Flags in Audio Evidence

Be suspicious when you see:

• Transcripts produced by investigators
• Claims of “absolute certainty”
• Novel methods without peer review
• Lack of documented procedures
• No acknowledgment of alternative interpretations
• Confidence without acoustic verification