What 911 Call Analysis Is and Why It's Still in Courtrooms

The prosecution's expert says the caller used the wrong words for an innocent person. No lab work, no error rate, no peer review. Just a checklist applied to a 911 call, and a jury that never hears the phrase "junk science."

That's the core problem with 911 call analysis. The technique, developed by retired Ohio police chief Tracy Harpster, claims that 20 "guilt" and "innocence" indicators can be read from 911 recordings. The entire method rests on a graduate thesis built from 100 recordings, according to Forensic Magazine. Harpster claims to have reviewed over 2,500 equivocal death cases and says he's unaware of any wrongful conviction resulting from the analysis, per Governing.

The scientific record tells a different story. Five independent studies have contradicted the method, including two from the FBI's Behavioral Analysis Unit, which found the indicators so inconsistent that some pointed in the opposite direction of Harpster's claims. The FBI initially promoted the work, then quietly reversed course, a reversal that went largely unnoticed by the agencies still buying his $3,500 training classes.

The institutional backlash has been blunt:

• The NACDL calls the practice "dangerous and insidious" and urges courts to add it to the junk pile of fake scientific theories.
• The Innocence Project identifies faulty forensic disciplines as the second most common factor in wrongful convictions, per ProPublica's investigation.
• The Colorado Bureau of Investigation has stated it would never advocate for 911 call analysis as a deciding factor in an arrest or conviction.
• North Carolina's Office of Indigent Defense Services has issued statewide warnings telling defenders to watch for the technique.

Here's what makes this dangerous in practice: the technique has appeared in more than 100 cases across 26 states, and prosecutors have learned to keep it out of admissibility fights entirely. A former Missouri prosecutor admitted in a 2016 email that the method wasn't recognized as science in her state, but that she was "getting creative … without calling it 'science'" to get it in front of juries. If the prosecution never frames the technique as scientific testimony, it may never face Daubert or Frye scrutiny. That means you can't count on the state to flag it. You have to recognize the indicators language yourself, usually buried in a detective's testimony or a report narrative.

The stakes are real. Russ Faria spent 3.5 years wrongfully imprisoned before a retrial where the 911 analysis testimony was cut short. Attorney Josh Tepfer, who has helped 288 wrongfully convicted people, now represents Jessica Logan in a case where the technique was used.

Your counter starts with the audio itself. Getting a precise, timestamped transcript of the call lets you test every claimed "indicator" against what the caller actually said, in context, and against the rest of your evidence. Tools like Evidence Copilot's 911 call transcript analysis give defense teams word-level timestamps and citations into the exact second of audio, so cross-examination questions are tied to quoted sources rather than memory. The same workflow applies when correlating a defendant's 911 statements with later jail calls, where accurate jail call transcription keeps the comparison honest.

The method has no scientific foundation. Your job is making sure the record proves it.

The Real Bottleneck: You Can't Challenge What You Haven't Heard

Start with the math. A public defender’s caseload runs 3 to 10 times the recommended amount, and every one of those files now arrives with hours of audio attached, according to reporting on the digital evidence volume crisis. A single felony case can mean a 911 call, two jail calls, and three bodycam files. You cannot challenge what you have not heard, and you cannot hear all of it by pressing play.

The scale is not theoretical. New York City alone assigned roughly 3.7 million 911 calls for an NYPD response in 2025, and Vera Institute data shows about 96% of those calls were not identified as violent crime. Caller language in those recordings reflects stress, confusion, or a health crisis far more often than anything a guilt indicator could measure.

Manual review collapses under that weight. Richmond Chief Public Defender Tracy Paner put it plainly in the ABA Journal: “I don’t have the luxury of time to necessarily watch 11 [videos], but I can definitely prioritize and watch three. And if I can pick the right three, it’s going to make all the difference.” That is triage, and it only works if the recordings are searchable first.

The workflow looks like this:

• Transcribe the 911 audio, jail calls, and bodycam with speaker diarization and word-level timestamps.
• Search the text for the terms the prosecution’s analyst flags, like “blood,” “sorry,” or “Huh?”
• Pull the timestamped segment, read the surrounding context, and compare it against the police report.
• Flag the inconsistencies and cite the exact second in your motion or cross-exam notes.

That last step is where the junk science defense gets built. A Forensic Magazine review documents five independent studies, including two from the FBI Behavioral Analysis Unit, that found Harpster’s indicators inconsistent or pointing the opposite direction. You cannot cross-examine on that record while listening blind. You need the transcript, the timestamp, and the quote in front of you.

For teams facing this volume, 911 call transcript analysis turns hours of audio into searchable, citable text, and jail call transcription lets you line up what your client said on the call against what the report claims. Every answer cites the exact moment it came from, and every AI output is marked for your review. The bottleneck is time, not effort. Triage is how you get it back.

Building the Counter-Narrative: From Transcript to Cross-Examination

Harpster's method treats a handful of words as guilt tells: saying "blood," answering "Huh?", failing to ask about the victim's condition. Each claim collapses the moment you read the caller's actual words in sequence instead of in isolation. Defense teams win these challenges with transcripts, not instincts.

The published record gives you plenty to work with. Five independent studies have contradicted the method, including two from the FBI Behavioral Analysis Unit, which found the indicators so inconsistent that some pointed the opposite direction. Harpster's list of 20 indicators traces back to a thesis built on 100 recordings. Meanwhile, ProPublica has documented the technique in more than 100 cases across 26 states, and the NACDL calls the practice "dangerous and insidious."

Build your counter-narrative in four passes:

• Test each indicator against context. Pull every flagged word from the timestamped transcript and read the surrounding lines. A caller who says "blood" while begging dispatch to hurry is not confessing; she is panicking.
• Cross-reference the 911 audio against the police report and bodycam. Report-versus-recording discrepancies are the raw material of impeachment.
• Correlate 911 language with later jail calls. Consistent accounts under stress cut against a guilt narrative; shifts between the two recordings give you questions.
• Quote the source, every time. A cross-exam question tied to a timestamp is far harder to dodge than one tied to an impression.

The volume problem is real: public defenders carry 3 to 10 times the recommended caseload, and 96% of 911 calls assigned to the NYPD do not involve violent crime. Searchable, timestamped transcripts turn that pile into a triage list. Evidence Copilot's 911 call transcript analysis and jail call transcription keep every claim tied to the second it came from.

Then frame the motion. Cite the FBI's reversal, the five failed studies, and the NACDL's position to argue the method fails Daubert as investigative lead dressed up as expertise. Prosecutors have admitted getting "creative" to avoid calling it science. Your motion should name that move directly.

A Working Review Workflow for 911 and Jail Call Evidence

You have the discovery dump, but finding the needle in 40 hours of audio is a slog. That is why your 911 call analysis workflow needs to move from passive listening to active text-searching.

Start by uploading the raw audio files directly into your case workspace. The system processes these files using Deepgram Nova, which generates diarized transcripts with word-level timestamps. You do not need to guess who is speaking. The software identifies the dispatcher and the caller, allowing you to rename speakers with one click. That change propagates across every segment instantly.

Once the transcripts are ready, you search for specific indicator keywords. The former police chief Tracy Harpster claims to identify "guilt" based on words like "blood" or "sorry" in 911 recordings. Independent studies have contradicted these claims, noting that such language often reflects stress rather than culpability. By searching for these terms, you can isolate the exact moments where the prosecution plans to build its case. You then listen to the specific clip, not the entire hour-long recording.

This triage step is critical when you are managing caseloads that are three to ten times the recommended amount. You cannot watch every video. You need to prioritize. When you find a discrepancy between the 911 call and the police report, the system flags it as a review signal. These signals appear in your activity feed before you even ask. They highlight potential inconsistencies between the audio and the written documentation.

To prepare for trial, you build an exhibit pack that cites the exact second of the audio. This pack includes video clips, still frames, and quoted excerpts with a citation index. Every AI-generated output is marked for human review, ensuring you retain control over the legal narrative. Tools like 911 call transcript analysis and jail call transcription help you correlate the initial call with subsequent communications. This allows you to show consistency or inconsistency in the defendant's narrative without manual scrubbing.

Your workflow should include these specific actions:

• Upload all audio and video files to create a unified case timeline.
• Search transcripts for keywords that match the prosecution's "guilt indicators."
• Review auto-generated signals that flag inconsistencies between reports and audio.
• Export a citation-ready exhibit pack for your motion to suppress or cross-exam.

The goal is to ground your arguments in verifiable, timestamped evidence. You are not relying on memory or vague recollections. You are pointing to the exact second where the caller’s tone shifts. This precision strengthens your position in the courtroom. It also saves you hours of manual review. The tool handles the tedious work of transcription and synchronization, so you can focus on the legal strategy.

Evidence Copilot provides this grounded evidence-review workspace for defense teams. Every AI answer cites the exact source moment it came from. You start with a 7-day full-access trial to test the workflow on your current case. No card is required. When you are ready, the Solo plan is $39 per user per month. This includes 15 media-processing hours. For public defender offices, the plan is $750 per attorney per year with 100 hours included. The pricing is transparent, and there are no feature tiers. You get the full workspace on every plan.

You can book a 20-minute walkthrough to see how the tool handles your specific discovery dump. Or, you can start on the Solo plan and upload your first file today. The process is simple. You upload, you search, you cite. That is the entire workflow. It removes the friction from evidence review. It gives you the leverage you need to challenge the prosecution’s narrative effectively.