What 911 Call Analysis Actually Is (and Why It's Classified as Junk Forensics)

Your client calls 911 for help. Two years later, a police officer tells a jury that the way she spoke on that call proves she's guilty. That's the practical effect of 911 call analysis, a technique that treats a caller's word choices and tone as evidence of deception, and it's now showing up in prosecutions across the country.

The technique traces back to the graduate research of a retired Ohio deputy chief, Tracy Harpster, who claims to have examined over 2,500 equivocal death cases. ProPublica documented more than 100 cases in 26 states where law enforcement applied his methods, often to assess the credibility, and even the guilt, of the callers themselves. The FBI helped spread the technique through its training channels.

The scientific foundation doesn't hold up. At least six peer-reviewed studies have failed to replicate the findings behind the training methodology, according to Cornell research on how callers become suspects. The FBI's own published data undercuts the core indicators: seasoned investigators agreed on whether a caller showed "voice modulation" in only 72% of 175 calls analyzed, and the FBI Law Enforcement Bulletin concluded these linguistic features should not be used on their own to evaluate veracity.

So why does it survive? Because it operates as what legal scholars call "shadow forensics". A Cornell Law Review article describes the technique as particularly pernicious because its underlying methods are never disclosed and its founders are never proffered as experts. Instead, police officers or 911 operators testify, which lets the analysis largely escape Daubert and Frye scrutiny entirely.

The pattern should look familiar. The same article compares its admission arc to bitemark evidence, another junk forensic discipline later linked to hundreds of exonerations. Faulty scientific disciplines are the second most common factor in wrongful convictions, per the Innocence Project. NACDL's president has called for adding 911 call analysis to the "junk pile of fake scientific theories."

For defense teams, the core angle is straightforward: force the technique into the light. That means demanding disclosure of the methodology, challenging the testifying officer's expert qualifications, and building a factual record that dismantles the specific indicators. When the prosecution leans on word frequency or tone, the counter is empirical. Over 80% of both innocent and guilty callers use the word "just," and one FBI follow-up study found the opposite correlation from the original research on voice modulation.

Practical review work comes first, though. Comparing the 911 audio against the police report, and against later jail calls, is where inconsistencies surface, and it's where a grounded transcript review pays off. Tools like Evidence Copilot handle 911 call transcript analysis and jail call transcription with word-level timestamps and speaker diarization, so you can cite the exact second the caller's account diverges from the officer's report. That record is what turns a junk-science argument into a cross-examination you can win.

How 911 Call Analysis Evades Daubert and Frye Scrutiny

A detective takes the stand and tells the jury your client's flat tone on a 911 call proves guilt. The jury hears it as science. No expert was disclosed. No methodology was tested. That's the admissibility gap at the heart of 911 call analysis, and it's why the technique survives in courtrooms that would exclude almost any other forensic claim under Daubert or Frye.

The Cornell Law Review calls this "shadow forensics." The label fits because the entire evidentiary structure is designed to stay hidden. Founders of the methodology are never proffered as experts. Instead, the officer who attended a one-day training session, or the 911 operator who listened to the call, testifies about what they "observed." That framing sidesteps the gatekeeping function of Rule 702 entirely.

The scale of the problem is documented. ProPublica's investigation found more than 100 cases across 26 states where the methods of former Deputy Chief Tracy Harpster were used to assess caller credibility. Many of the agencies involved were unaware their own personnel had been trained in the technique until shown documentation by reporters.

Three structural features let 911 call analysis evade scrutiny:

• No expert disclosure. The person who developed the methodology never testifies, so there's no methodology to challenge under *Daubert*.
• Lay witness framing. Officers describe "suspicious behavior" as personal observation, not forensic opinion, which sidesteps expert disclosure requirements.
• Untested underlying claims. At least six peer-reviewed studies failed to replicate the findings of the original training methodology, yet the technique persists.

The replication failures matter because they show the foundation is hollow. Cornell researchers found that demeanor-based suspicion, especially the absence of "expected" emotion, is a flawed predictor of guilt. Four out of five study participants cited emotion as the factor that most influenced their suspicion of a caller, even though trauma responses vary widely across personality, culture, and neurotype.

Defense teams who want to challenge this testimony need to do the work the prosecution skipped. That means getting a clean transcript of the 911 audio with speaker labels and timestamps, then comparing the caller's actual words to the officer's characterization on the stand. A defender running 911 call transcript analysis through a grounded evidence workspace can pull the exact second the officer claims to have heard "flat affect" or "evasive phrasing" and check it against the audio. When the transcript contradicts the testimony, you have a cross-examination roadmap and, in many cases, the foundation for a motion to exclude the officer's opinion as an undisclosed expert.

The same principle applies to subsequent calls. If your client made statements on a recorded jail call that contradict the prosecution's narrative about the 911 call, jail call transcription with diarization lets you line up both recordings on a single timeline and show the jury where the story changed, and whether the change came from coaching or from the original call being mischaracterized.

The takeaway for your next motion hearing: shadow forensics only stays in the shadows if nobody shines a light on it. Force disclosure. Name the methodology. Cite the replication failures. And make sure the transcript you're working from is precise enough to contradict whatever the officer remembers on the stand.

The Science on Voice Modulation, 'Just,' and Demeanor Is Contradictory

When the state's witness testifies that your client's flat tone "indicates guilt," the most powerful cross-examination weapon is the prosecution's own literature. The empirical record on 911 call analysis doesn't just have gaps; it has direct contradictions between its foundational studies.

Take voice modulation. The original 2009 study by Harpster, Adams, and Jarvis found that limited tone or pitch change correlated with guilt. But a follow-up study published in the FBI Law Enforcement Bulletin found the opposite: homicide offenders were more likely to exhibit voice modulation. In staged-suicide versus true-suicide calls, voice modulation didn't distinguish callers at all. Same technique, opposite findings.

The inter-rater reliability data is worse. When seasoned investigators with 10+ years of experience evaluated 175 calls, they agreed on whether a caller even demonstrated voice modulation in only 72% of them. That means trained observers disagreed roughly one time in four about whether the phenomenon existed before anyone debated what it meant. If the observation itself is that unstable, any conclusion built on it collapses on cross.

The word "just" tells a similar story. Proponents cite it as a minimization marker signaling involvement, but the FBI data shows over 80% of both innocent and guilty callers used it. The frequency gap (4 uses per call for involved callers versus 2.5 for uninvolved) depends on counting word occurrences in a single call, a distinction your client's natural speech patterns can erase entirely.

The demeanor evidence is equally shaky. Cornell research found that four out of five participants cited the caller's emotion as the factor that raised or lowered their suspicion, yet at least six peer-reviewed studies failed to replicate the findings behind the training methodology. As researcher Jessica Salerno puts it, having expectations for demeanor on a 911 call is dangerous because of cultural, personality, and neurodivergent differences. There is no "correct" amount of grief.

Even the FBI's own authors concede the point. They conclude that "the evidence on voice modulation and minimization as appropriate indicators of deception in 911 calls is mixed" and that these features should not be used on their own to evaluate veracity.

For cross-examination, the contradictions give you a clean sequence:

• Establish the witness relies on voice modulation, then confront them with the 2021 finding that offenders modulated more.
• Ask whether they can identify voice modulation consistently, then present the 72% agreement rate.
• Ask if "just" indicates guilt, then note that over 80% of innocent callers use it too.
• Close on the FBI authors' own concession that these features cannot stand alone.

Getting there requires an accurate transcript with word-level timestamps, because the counting exercise only works if the counts are right. When we review 911 call audio through Evidence Copilot, every quoted word carries a citation back to the exact moment in the recording, so the impeachment numbers hold up when opposing counsel checks them. The same workflow applies when comparing the call against later jail call transcription, where shifts in narrative often reveal coaching rather than guilt.

A Defense Workflow: Turning the 911 Call Into Your Best Impeachment Source

If the prosecution is leaning on 911 call analysis, your best weapon isn't a hired expert. It's the call itself, worked over carefully and compared against everything else in discovery.

Start with the raw audio. Don't rely on the dispatcher's summary or the officer's report; request the original recording file from the agency. Then run a proper 911 call transcript analysis with speaker diarization, so you know exactly who said what and when. Word-level timestamps matter here, because the prosecution's theory often turns on a single phrase, like how many times the caller said "just." That's not a hypothetical: the FBI's own research on 911 homicide calls found that over 80% of both involved and uninvolved callers used the word "just," which makes any argument built on word frequency fragile.

Once you have a diarized transcript, put it side by side with the police report and any jail calls. The goal is to map every discrepancy: details the caller gave dispatch that the responding officer recorded differently, or narrative shifts between the 911 call and later statements. This is where a grounded review workspace like Evidence Copilot earns its keep, because every comparison is cited back to the exact second of audio or page of the report, and auto-signals flag report-versus-call inconsistencies before you have to hunt for them.

Pay special attention to what the call captures beyond the caller's words. Background voices, mutterings, and ambient audio have high investigative value, and they're routinely ignored in a prosecutor's analysis. A few things worth flagging in your review:

• Diarization errors: if the transcript misattributes a line to the caller, the state's "suspicious statement" may not be the caller's at all.
• Background audio that contradicts the timeline in the police report.
• Tone claims you can test: seasoned investigators agreed on whether a caller showed voice modulation in only 72% of 175 calls analyzed, so an officer's ear is not a scientific instrument.
• Jail calls, transcribed and searchable, that show whether the client's account stayed consistent or shifted under pressure.

This workflow is the antidote to what Cornell's law review has called shadow forensics: techniques that evade Daubert scrutiny because the methods are never disclosed and the founders never take the stand. You can't cross-examine a hidden methodology, but you can put the underlying audio on the record. When the state's theory rests on demeanor, remember that Cornell research found 4 out of 5 study participants judged callers by emotion, a factor researchers call scientifically unsupported given cultural and neurodivergent differences.

Finally, compare the 911 call against jail call transcripts to trace the narrative's evolution. Consistent jail call transcription and call-by-call comparison can show whether investigators locked onto the caller early and then read every later statement through that lens. That's confirmation bias, documented in your own discovery, and it plays in cross-examination far better than an abstract argument about junk science ever will.

Cross-Exam and Suppression Moves That Actually Work

The strongest cross-exam moves start with a single question to the testifying officer: what method are you applying, and where was it published? In a Cornell Law Review analysis, Christopher Lau describes 911 call analysis as "shadow forensics" precisely because underlying methods stay undisclosed and the technique's founders are never proffered as experts. That gap is your opening.

Force the choice. Either the prosecution discloses the methodology behind the suspicion, or the officer must qualify as an expert, which triggers Daubert scrutiny of a technique that has never survived it. ProPublica documented more than 100 cases across 26 states built on the methods of former Deputy Chief Tracy Harpster, and roughly 50 agencies canvassed on the practice produced only 11 responses, with some unaware their own personnel were involved.

Demeanor-based suspicion is the next target. Cornell research found that 4 out of 5 participants treated a caller's emotion as the factor that raised or lowered suspicion, yet at least six peer-reviewed studies failed to replicate the police training findings behind it. Jessica Salerno, an associate professor at Cornell, puts it plainly: expecting a particular demeanor on a 911 call is dangerous because culture, personality, and neurodivergence all shape how people sound under stress.

Then use the FBI's own words. In the Bureau's Law Enforcement Bulletin, authors reviewing voice modulation and minimization in 911 homicide calls concluded the evidence is mixed and those linguistic features should not be used on their own to judge veracity. The same article reports that over 80% of both innocent and guilty callers used the word "just," and that seasoned investigators agreed on voice modulation in only 72% of 175 calls reviewed.

• Demand the training materials, validation studies, and error rate behind any suspicion the officer attributes to the call.
• Ask whether the officer applied a checklist before or after learning the case facts, which exposes confirmation bias.
• Pin down demeanor testimony with the Cornell and FBI findings before the jury hears the word "deceptive."
• Paper the record with your own transcript, because a word-level 911 call transcript analysis lets you quote the call at the exact second the officer mischaracterizes it.

What makes these moves work is the record you build before the hearing. When the call audio and the police report diverge, the inconsistency is the argument, and a grounded transcript with word-level timestamps is what lets you put both versions side by side and cite the moment. Evidence Copilot was built for exactly this kind of review, and teams that pair jail call transcription with the 911 audio can track how a client's account shifts across a case, then decide whether that shift reflects coaching or the ordinary stress the research describes.