What 911 Call Analysis Actually Is, and Why It's in Your Case

You've got a 911 call in discovery, and the officer who testified about it never claimed to be an expert. That's not an accident. It's the whole design.

911 call analysis is a technique invented by Tracy Harpster, a former Ohio deputy police chief, who built it from a 2006 master's thesis reviewing 100 homicide calls. He later expanded it with FBI analyst Dr. Susan Adams into 200 calls and 38 indicators, packaged as the proprietary "Considering Offender Probability in Statements" (COPS) Scale, according to a University of Cincinnati Law Review analysis. The method has since appeared in more than 100 documented cases across 26 states, per ProPublica's investigation.

The indicators are the tell. Watch for an isolated "please," a call that opens with "hi," a flat "Huh?", or the absence of an immediate plea for help. If you see those flagged in a police report or hear an officer describe a caller as "too calm," someone in that case sat through Harpster training.

Here's the evasion problem. A 2026 Cornell Law Review article calls it "shadow forensics": the underlying method is never disclosed, and its creators never take the stand as experts, so a dispatcher or detective can testify about caller culpability as a lay witness after attending a class. Harpster himself has consulted on more than 2,500 homicide investigations and has never testified as an expert witness in a criminal trial, per the UC Law Review. Training runs up to $3,500 for an eight-hour class, often taxpayer-funded, Forensic Magazine reports.

The science doesn't hold. At least eight studies have failed to replicate the method, per the Innocence Project, including two FBI Behavioral Analysis Unit studies whose authors warned some indicators pointed in the opposite direction. Dr. Jessica Salerno's research suggests listeners are detecting an absence of expected emotion, not deception. Robert Roberson's flat affect, read as guilt, was later explained by an autism diagnosis.

The backlash is real and it's 2026-shaped. NACDL has condemned the practice, North Carolina's Office of Indigent Defense Services issued a statewide warning, and the Illinois Supreme Court agreed to review Jessica Logan's conviction after a Harpster-trained detective testified about her call. Russ Faria served 3.5 years before acquittal on retrial; Riley Spitler's Michigan conviction was overturned. Harpster disputes the framing, saying the analysis has been used to exonerate people, and ProPublica notes it never reported the technique caused an innocent person's conviction.

For your transcript review, this changes what you're looking for:

• Pull the call audio and get a timestamped, speaker-attributed transcript so you can quote the exact words an officer characterized.
• Ask every testifying officer whether they attended Harpster training, then object under FRE 701 and 702.
• Check whether the state proffered anyone as an expert, or slid the method in through lay testimony.
• Cross-reference the call against the report and bodycam for inconsistencies.

Evidence Copilot's 911 call transcript analysis gives you diarized, word-level timestamped transcripts with clickable citations into the exact moment, and jail call transcription works the same way when the custody calls matter too. Every AI output is marked for human review, and originals stay immutable.

Why the Science Doesn't Hold Up

You need to know that the science behind 911 call analysis has failed repeatedly. The method claims to detect guilt through specific verbal cues, yet it lacks the empirical foundation required for forensic evidence.

At least eight independent scientific studies attempted to replicate the original findings and found the indicators do not hold up under scrutiny. Research from the Innocence Project highlights that these replications included two separate studies by the FBI Behavioral Analysis Unit. In those tests, FBI experts warned that the method may exacerbate bias and that some indicators actually pointed in the opposite direction of what the original theory predicted.

Dr. Jessica Salerno, a psychology professor at Cornell University, identified what observers are actually detecting. She found that listeners are reacting to the absence of expected emotional display, not to genuine signs of deception. For police officers, even moderate emotion was not enough to reduce suspicion; callers had to sound highly distressed before perceptions shifted. This explains why the technique often misidentifies neurodivergent individuals.

Consider the case of Robert Roberson, who spent more than two decades on Texas death row. Investigators interpreted his flat affect during a 911 call as a marker of guilt. His later autism spectrum diagnosis provided a clear alternative explanation for that behavior, yet the initial analysis had already shaped the investigation against him.

This reliance on unvalidated indicators creates a dangerous feedback loop. The University of Michigan STPP policy brief notes that the technique routinely reinforces investigative confirmation bias. When an investigator believes a caller is guilty based on a "please" or a "Huh?", they divert attention from more reliable physical evidence. This contamination affects interrogations, forensic interpretation, and which leads get pursued.

To challenge this effectively, you must pinpoint the exact words and timestamps the officer is referencing. Vague characterizations are hard to impeach, but specific audio segments are not. A 911 call transcript analysis allows you to verify the officer’s claims against the actual audio record. You can see if the "indicators" they cite actually exist or if they are embellishments.

Similarly, if the case involves post-arrest statements, jail call transcription provides the precise language used. This ensures you are not relying on memory or summary reports when preparing your cross-examination.

The failed replications are not just academic footnotes. They are your primary defense tool. When an officer testifies that a caller sounded guilty, you can point to the body of scientific evidence that contradicts their training. You can argue that the technique is not accepted in the scientific community and that its use introduces prejudice rather than probative value.

By grounding your objections in these specific studies, you shift the burden back to the prosecution. They must prove the method’s reliability, not just its popularity. The data shows it fails that test. Your job is to make sure the jury never hears the flawed interpretation without first hearing the scientific rebuttal.

How to Challenge It: FRE 701 vs. 702 and the Lay-Testimony Problem

The whole fight usually turns on one question: is the officer testifying as a lay witness or an expert? Prosecutors know the difference matters, which is why the technique almost never arrives labeled as science. Instead, a detective who sat through an 8-hour training class (one that can cost taxpayers up to $3,500) takes the stand and walks the jury through the caller's word choices as if reading a weather map. The University of Cincinnati Law Review calls this the technique's defining evasion: Harpster's method gets "mixed into in-court statements and lay testimonies to sneak it past the standards for expert testimony."

Here's the rule structure you're working with. Under FRE 701, a lay witness may offer opinions rationally drawn from perception, helpful to the jury, and not based on specialized knowledge. The moment a witness starts applying the COPS Scale's 38 indicators, saying "Huh?" or an isolated "please" marks guilt, they're using specialized knowledge, which pushes them into FRE 702 and Daubert territory. Under Daubert's five factors (testability, peer review, error rate, standards, and general acceptance), the method collapses: at least eight scientific studies failed to replicate its claims, and two FBI Behavioral Analysis Unit studies found some indicators ran "in the opposite direction of what was previously found." In Frye states, the general-acceptance showing fails for the same reason. Christopher Lau's 2026 Cornell Law Review article names the pattern "shadow forensics": the founders never testify as experts, the underlying data is never disclosed, and the technique escapes scrutiny entirely. Harpster himself has consulted on over 2,500 homicide investigations without ever testifying as an expert in a criminal trial.

So your job is to force the technique out of the shadows and onto the record. Build your cross around three questions:

• Where did you train? Ask whether the witness attended Harpster's course, paid for it, and whether the COPS Scale materials were produced in discovery. Training attendance converts a "lay" witness into a 702 problem.
Are you offering a culpability opinion? Pin the witness down: does "too polite" mean the caller is the offender? If yes, that's an opinion on guilt, not a perception-based lay observation.Where's the method? Demand the underlying study. Harpster and Adams never released how they obtained their call sample or made their data public, so there is nothing for your expert to test.What does the transcript actually say? Get the timestamped, diarized call in front of the witness and test each claimed indicator against the exact words spoken.Whose expectation of emotion? Dr. Jessica Salerno's research suggests observers are reacting to absent emotional display, not deception, and that officers stay suspicious "even [with] moderate emotion."

The case law gives you cover. Courts have rejected Harpster himself as an expert (Alabama, 2009) and excluded a Harpster-trained detective in Nevada, and Michigan's Riley Spitler murder conviction was overturned because 911 call analysis and similar statement-analysis techniques should never have been admitted, per the UC Law Review analysis. Salerno's advice to the defense bar is blunt: learn to identify and challenge attempts to introduce the technique through lay testimony. That challenge starts with a clean transcript. When you're working a call for cross, timestamped speaker attribution and word-level citations matter, which is why tools built for 911 call transcript analysis (and jail call transcription) anchor every quote to the exact second of audio, so no one can argue later about what the caller actually said.

Documented Cases Where Challenges Worked

The clearest evidence that 911 call analysis can be challenged isn't in a lab study. It's in the case files of four people who lost years of their lives to it and got them back.

Russ Faria spent 3.5 years in prison before his acquittal on retrial, in a case where testimony about his 911 call played a part, according to Forensic Magazine's investigation. Huwe Burton was arrested at 16 after calling 911 to report his mother's death, and he walked out in 2019 after more than 20 years behind bars, a case the Innocence Project documented as an exoneration where flat affect had been read as guilt.

Riley Spitler's Michigan murder conviction was overturned because 911 call analysis and other statement-analysis techniques should never have been admitted, as the University of Cincinnati Law Review notes alongside cases where courts rejected both Harpster himself (Alabama, 2009) and a Harpster-trained detective (Nevada). And Jessica Logan's case is still live: the Illinois Supreme Court agreed to review her conviction after a Harpster-trained detective testified about her call, per ProPublica's reporting.

What did the winning challenges have in common? A few things worth studying before you draft your own motion:

• Identifying the witness's training. The technique almost always arrives through lay testimony from a detective or dispatcher who took Harpster's 8-hour class, not through a qualified expert. That's the FRE 701 vs. 702 wedge.
• The actual call audio and a reliable transcript. An officer's paraphrase of "what the caller said" is where the analysis lives. Without word-level timestamps and speaker attribution, you're arguing about memory instead of the record.
• The scientific record. At least eight studies failed to replicate the method's claims, including two FBI Behavioral Analysis Unit studies that found some indicators ran in the opposite direction.
• The methodology's opacity. Harpster and Adams never released their call data or explained how they obtained it, which makes cross-examination on reliability nearly impossible for the prosecution to survive.

Be honest with yourself about the counterarguments, because the prosecution will make them. Harpster maintains he has "never" seen an innocent person convicted using the analysis and says it has exonerated people. Adams argues no single indicator determines guilt, only indicators examined "in combination." ProPublica itself has acknowledged it never reported that the technique caused the conviction of innocent persons. Your challenge is stronger framed as "this method is scientifically unvalidated and inadmissible" than as "this method convicted an innocent man."

That framing also changes what you need from discovery. You need the call audio, a defensible transcript, and the training records, and you need them organized early enough to spot the Harpster indicators ("Huh?", the isolated "please," opening with "hi") before the witness takes the stand. When we built Evidence Copilot's workflow around 911 call transcript analysis, the goal was exactly that: timestamped, diarized transcripts where every claim about the call links back to the exact second of audio. The same holds for jail call transcription, where the same statement-analysis logic sometimes shows up in a second disguise.

Faria, Burton, Spitler, and Logan prove the challenges work. Your job is to build the record that makes yours one of them.

Building the Record: Transcript Review Workflow for 911 and Jail Calls

When a Harpster-trained detective takes the stand and says your client "showed no plea for help" or sounded "too polite," the only thing that beats a characterization of the call is the call itself. Your job before that cross-exam is to build a record precise enough that the jury can read the exact words, at the exact second, and decide for themselves.

Start with a diarized, timestamped transcript of every 911 recording and every jail call in discovery. A Wisconsin Innocence Project co-director writing in the Cornell Law Review calls 911 call analysis "shadow forensics" precisely because the method hides inside lay testimony. The counter is literal: pull the detective's claimed indicator, put the actual transcript segment beside it, and show the court what was really said.

Your transcript review workflow should produce three things:

• Word-level timestamps so any excerpt you quote in a motion or cross can be verified against the audio to the second.
• Speaker attribution that survives scrutiny: caller, dispatcher, background voices, and on jail calls, the recorded party versus the client. Misattributed lines are the fastest way to lose a credibility fight.
• A quote bank keyed to the named indicators: "Huh?", an isolated "please," opening with "hi," "blood" versus "bleeding". When the officer testifies, you already have the actual utterance, timestamped, in front of you.

Jail calls deserve the same discipline. Prosecutors love to paraphrase weeks of calls into a single sentence about "consciousness of guilt." A full transcript lets you show the context the paraphrase dropped, and a diarized record tells you who actually said what. Tools built for defense teams, like Evidence Copilot's 911 call transcript analysis and jail call transcription, generate diarized transcripts with word-level timestamps and let you rename speakers once, across every segment, so the record stays consistent from preliminary hearing through trial.

The stakes justify the effort. ProPublica has documented more than 100 cases across 26 states involving Harpster's methods, and Russ Faria spent 3.5 years in prison partly on 911 call analysis testimony before an acquittal on retrial. The University of Michigan's policy brief notes the technique reinforces investigative confirmation bias and has contributed to wrongful convictions. Your transcript record is how you make that visible to a judge.

Keep every AI-generated transcript marked as pending human review, and verify the critical excerpts by ear before you quote them. The tool builds the record; you certify it.

If you want to see this workflow on one of your own cases, book a 20-minute walkthrough at evidencecopilot.com/book, or start on the Solo plan at evidencecopilot.com/signup and upload your first 911 call.