What 911 Call Analysis Claims (and Why It Keeps Failing Science)

If a detective tells your jury that a caller "didn't sound right," you need to know exactly where that opinion came from. The answer is almost always 911 call analysis, a technique built on a 2006 master's thesis, never validated, and now facing calls for bans from judges, prosecutors, and the Innocence Project alike.

The origin story matters because it sets the evidentiary ceiling. Tracy Harpster developed the COPS Scale as his thesis at the University of Cincinnati, examining just 100 calls. It was exploratory research only, and one co-author later told journalist Brett Murphy the findings were "not ready for real-world application." The scale kept growing anyway: 16 factors in 2006, 18 indicators in a 2009 paper, and 38 indicators across 200 calls by the 2017 book, with no independent validation at any step.

Then came the replications. At least eight subsequent scientific studies attempted to test the method, and every one found the supposed guilt indicators don't hold up under scrutiny, according to the Innocence Project. Even the FBI's own January 2022 Law Enforcement Bulletin concluded that voice modulation and minimization "should not be used on their own for evaluating veracity versus deception." That same FBI article found trained authors with a decade-plus of investigative experience could only agree on whether voice modulation existed in 72% of 175 calls. If the raters can't agree the indicator is present, it can't measure anything.

The FBI authors also undercut the specific "markers" prosecutors lean on. Across the same 175 calls, more than 80% of both guilty and innocent callers used the word "just," and the frequency gap was small: offenders averaged 4 uses versus 2.5 for uninvolved callers.

Before you walk into a hearing, anchor your record on these points:

• The original study was a 100-call master's thesis, exploratory by design, with a co-author disclaiming real-world use.
• At least eight independent studies failed to replicate the claimed indicators.
• The FBI's own publication disclaims using these linguistic features as standalone deception measures.
• Low emotional expression, the behavior most often read as "guilt," correlates with trauma, culture, neurodivergence, and shock regardless of whether the caller was later convicted or exonerated.

The practical problem is that prosecutors rarely proffer Harpster as an expert. They put a trained detective on the stand and let the guilt inference ride in as lay opinion. When you're pulling the 911 audio, the interrogation footage, and the detective's report into one review file, tools like Evidence Copilot keep every quote tied to the exact second of audio, which is what a 911 call transcript analysis challenge actually requires. The same timestamped workflow applies to jail call transcription when the prosecution's narrative spans multiple recordings.

Know the baseline before the hearing starts: the technique fails testability, error rate, standards, peer review, and widespread acceptance. Every Daubert prong, documented in print.

The Shadow Forensics Workaround: How Prosecutors Sneak It In

Prosecutors rarely put 911 call analysis on the stand as expert testimony. They don't have to. Instead, they hand the technique to a trained detective or dispatcher and let it walk in through the front door as lay opinion.

Christopher Lau, Co-Director of the Wisconsin Innocence Project, calls this "shadow forensics" in a 2026 Cornell Law Review article: techniques whose underlying methods are never disclosed and whose founders are never proffered as experts, so they largely escape the Daubert or Frye scrutiny that expert testimony would trigger. The structure is deliberate. Jurors hear an officer describe a caller's word choice or flat tone as a guilt indicator without ever learning that the observation traces back to a disputed methodology.

The playbook is documented. Tracy Harpster, who developed the technique from a 2006 master's thesis built on just 100 calls, circulated a handout offering "ten tips for prosecutors to use 911 call analysis as part of their prosecution strategy without needing to seek the required qualification for expert testimony." At least eight subsequent studies failed to replicate the method's findings, and the Innocence Project has flagged the workaround as a core reason the technique survives in courtrooms at all. More than 100 cases across 26 states have been documented.

The interrogation room is the part defenders miss. In a 2012 Fremont, Ohio case, Detective Jason Kiddey used a 911 call analysis handout during questioning, and the suspect cracked, pleading to manslaughter with an 11-year sentence. That's a separate constitutional problem from trial admissibility, and it's worth pulling the full custodial record to see whether the technique shaped the questioning.

What to hunt for in discovery:

• Detective or dispatcher testimony describing a caller's tone, word choice, or "minimization" as suspicious.
• Training certificates, seminar handouts, or COPS Scale materials in the officer's file.
• Interrogation transcripts where an investigator references the call analysis to pressure a statement.
• Any mention of Harpster, call analysis training, or "equivocal death" language in reports.

A 911 call transcript analysis that pairs the audio against the responding officer's report often surfaces exactly where the inference got inserted. The same applies when you run jail call transcription alongside interrogation footage, since the two records together show whether the technique migrated from the phone call into the questioning.

The Confirmation Bias Trap: What 'Low Emotion' Actually Means

When a detective takes the stand and says your client "sounded too calm" on the 911 call, the jury hears a scientific-sounding conclusion. 911 call analysis has trained them to hear it that way. But the research behind that inference tells a very different story.

Dr. Jessica Salerno's psychology research at Cornell University, highlighted by the Innocence Project, found that what investigators read as "guilt" is really just low emotional expression. Both civilians and police officers became significantly more suspicious of callers who expressed little emotion. For police officers, even moderate emotion wasn't enough to lower suspicion; callers had to sound highly emotional before officers' perceptions shifted.

Here's the part that matters for your cross-examination: that pattern showed up regardless of whether the caller was later convicted or later exonerated. The "guilty caller" signal isn't detecting guilt at all. It's detecting flat affect, and flat affect has innocent explanations:

• Trauma responses that blunt emotional expression in acute crisis
• Cultural norms around public displays of emotion
• Personality and temperament differences
• Neurodivergence, including autism and alexithymia
• Shock and dissociation in the immediate aftermath of violence

Salerno's warning about the downstream damage is worth quoting in motions: confirmation bias "does not stay confined to that single interaction. It also shapes the interrogations that follow, influences forensic interpretation, narrows investigative focus, and can determine which leads are pursued and which are ignored." Once an officer decides your client sounded guilty, the investigation bends around that conclusion.

Even the FBI's own research undercuts the technique. A January 2022 FBI Law Enforcement Bulletin article found that across 175 calls, trained FBI authors with 10+ years of investigative experience, working with a forensic psychologist, could only agree on whether a caller demonstrated voice modulation in 72% of cases. If seasoned FBI analysts disagree more than a quarter of the time, a detective who sat through an eight-hour training has no business offering lay opinions about your client's tone. The Bulletin's own conclusion: voice modulation and minimization "should not be used on their own for evaluating veracity versus deception."

The practical move is to build this record early. Get a timestamped, diarized transcript of the call, then line up every "he sounded calm" observation against the actual audio. Evidence Copilot's 911 call transcript analysis gives you word-level timestamps with clickable citations, so you can impeach a characterization with the exact second it contradicts. The same workflow applies when the state later tries the same trick with recorded jail calls, where jail call transcription with speaker labels keeps who-said-what straight.

Frame the argument simply: the state is asking the jury to convict on an emotional read that the research shows is a bias, and that the FBI's own analysts can't reliably reproduce.

How to Challenge 911 Call Analysis at Trial and on Appeal

The prosecutor won't call an expert. That's the whole trick. Your job is to make the judge see that the detective testifying about "guilt indicators" is an expert in disguise.

Start with the motion. If the state proffers a detective or dispatcher who trained in 911 call analysis and will testify about the caller's word choice, cadence, or emotion, move in limine to exclude under FRE 701. Lay opinion must be rationally based on perception and cannot rest on scientific, technical, or specialized knowledge within the scope of FRE 702. A detective who took an eight-hour, taxpayer-funded course (programs cost up to $3,500) is applying a methodology, not sharing a perception. The University of Cincinnati Law Review lays out this FRE 701/702 framework in detail.

If the court treats the testimony as expert opinion, run the Daubert factors and make the state carry each one:

• Testability: at least eight independent studies have tried to replicate the technique and all failed to validate its indicators.
• Peer review and publication: the COPS Scale grew from 16 factors in the 2006 thesis to 38 indicators in the 2017 book, with no validated methodology disclosed along the way.
• Error rate: none exists. Even the FBI's own authors, with 10+ years of investigative experience plus a forensic psychologist, agreed on whether voice modulation was present in only 72% of 175 calls.
• Standards and acceptance: no controlling standards, and the technique is rejected in jurisdictions including Alabama and Nevada.

Force disclosure of the underlying method. Christopher Lau's 2026 Cornell Law Review article names the pattern "shadow forensics": the founders are never proffered as experts, the methods are never disclosed, and a trained officer testifies instead. Demand the training materials, the scoring instrument, and the basis for every claimed indicator. When the state can't produce a validated protocol, the record you build supports both the trial objection and later post-conviction relief.

Use the government's own research as impeachment. The FBI's January 2022 Law Enforcement Bulletin concluded that voice modulation and minimization "should not be used on their own for evaluating veracity versus deception." Dr. Jessica Salerno's research shows officers read low emotional expression as guilt, and that trauma, culture, personality, neurodivergence, and shock all shape crisis responses, regardless of whether the caller was later convicted or exonerated. Cross the detective on both.

Check the interrogation room, too. In a 2012 Fremont, Ohio case, a detective used a call analysis handout to pressure a suspect into pleading to manslaughter. If the technique surfaced in your client's custodial interrogation, evaluate a voluntariness challenge.

For post-conviction work, you're not alone. The Exoneration Project and Center for Integrity in Forensic Sciences have taken these cases, the Innocence Project has documented the science, and North Carolina's Indigent Defense Services has warned attorneys statewide to watch for it. Tools like Evidence Copilot can help you build the record, pairing a diarized 911 call transcript analysis with jail call transcription so every quoted "indicator" is tied to the exact second of audio. The Michigan policy brief makes the closing point for you: the technique reinforces confirmation bias and diverts attention from more reliable evidence.

Building the Record: Transcripts, Timelines, and Cross-Exam Prep

None of the shadow-forensics arguments in the last section matter unless you can put the actual words in front of the court, timestamped and quotable. Building that record is unglamorous work, and it's where most 911 call challenges are won or lost.

Start with the transcript itself. Get a timestamped, speaker-diarized transcript of the 911 call, not a dispatcher's summary. Word-level timestamps let you pin every quoted phrase to the exact second, which matters when you're cross-examining a detective about what the caller "actually said" versus what the detective's training told him it meant. The same goes for the related jail call transcription: in the 2012 Fremont, Ohio case, a detective used a 911 call analysis handout during interrogation, and the suspect "cracked" and pled to manslaughter with an 11-year sentence, so knowing what was said in the interrogation room and the jail phone calls tells you whether the technique was deployed as a pressure tool before trial. If it was, that's a voluntariness question, not just an admissibility one.

Next, map the call against everything else. Lay the 911 audio timeline over the police report and the bodycam footage, moment by moment. You're looking for the seams: the report describes the caller as "calm and matter-of-fact," but the audio shows a caller screaming; the officer's report says the dispatcher flagged the call as suspicious, but the dispatch log shows no such notation. Those seams are your cross-exam. Dr. Jessica Salerno's research found that both civilians and police officers became significantly more suspicious of callers who expressed low emotion, and the pattern held regardless of whether the caller was later convicted or exonerated, so an officer's "something felt off about the caller" testimony needs to be tied to specific, quotable moments you can play back in court.

Then draft the cross from the source material, not from memory. Every question should trace back to a quoted moment:

• "You testified the caller sounded 'controlled.' Play the 2:14 mark. Is this what you call controlled?"
• "Your training was eight hours of instruction. Is that correct?"
• "You didn't review the caller's trauma history before forming your opinion, did you?"
• "The FBI's own Law Enforcement Bulletin says these indicators shouldn't be used on their own. Were you aware of that?"

That last question lands because the FBI's January 2022 article concluded that voice modulation and minimization evidence "should not be used on their own for evaluating veracity versus deception," and even trained FBI authors with a decade of investigative experience could only agree on whether voice modulation was present in 72% of calls across 175 homicide calls. That's your error-rate question, handed to you by the government's own publication.

This is the workflow that makes the shadow-forensics challenge concrete: a grounded 911 call transcript analysis gives you the timestamps, the diarized speakers, and the quotable moments, and a parallel jail call transcription covers the interrogation side. Tools like Evidence Copilot are built around this pattern, generating timestamped transcripts and cross-exam drafts where every suggested question ties back to a quoted source moment. The transcript is the record. The cross-exam is what you do with it.