Why the Interrogation Transcript Is Your Biggest Discovery Bottleneck

Public defenders carry caseloads three to ten times the recommended amount, and the discovery keeps growing. Each officer's body camera now captures roughly 32 files, 7 hours, and 20GB of video every month, according to reporting on AI transcription in public defense. Interrogation footage, 911 audio, and jail calls pile on top of that, and much of it never gets watched.

That gap is where cases quietly go wrong. Harris County public defender Jacquelyn Carpenter put it plainly: defenders used to rely on the truth of the offense report, and now they can see that the report and the recording don't always line up. In one arson prosecution, bodycam evidence led to outright dismissal. A client whose interrogation statement sits unreviewed, or whose 911 call contradicts the report, absorbs a cost that better review would have caught.

The bottleneck is structural, not personal:

• Caseloads at 3-10x recommended levels leave no room for full video review.
• Volume per officer keeps climbing: 32 files and 7 hours monthly at 720p.
• Some public defenders average 7 minutes per case.
• Offense reports and footage frequently diverge, and nobody catches it.

Technology has started to close the gap. Harris County Chief Public Defender Alex Bunin reports roughly a 50% reduction in review time with AI transcription, and a Nevada investigator cut a 21-video review from 19 hours to 2. Those gains matter most when the tooling produces something you can cite: timestamped transcripts, speaker labels, and answers that point back to the exact moment in the recording. Evidence Copilot builds review around that principle, with every AI output marked for human review and originals left untouched.

Transcripts are the entry point, not the finish line. Whether you're pulling a statement from interrogation footage or running 911 call transcript analysis alongside it, the work only pays off when the record is searchable, citable, and checked against everything else in the file.

The 2026 Legal Landscape: Figueroa, Portillo, and Foreign-Language Interrogations

If your client gave a Spanish-language statement to police and the Commonwealth never handed you an English transcript, you might assume that statement stays out of the trial. In Massachusetts, as of August 2026, you'd be wrong about half of that assumption.

Commonwealth v. Figueroa, No. 25-P-1027 (Mass. App. Ct. Aug. 7, 2026), affirmed an assault-and-battery conviction where the interrogating detective impeached the defendant with prior Spanish-language statements, even though the Commonwealth never produced an English police interrogation transcript for the defense. The Appeals Court held the trial judge had discretion to allow it. The stakes were real: the charge carried up to 10 years in state prison or 2.5 years in the house of correction, and a $5,000 fine, per one defense-firm analysis.

The baseline rule comes from Commonwealth v. Portillo, 462 Mass. 324 (2012). A judge may require an English transcript before trial and may exclude foreign-language statements from the case-in-chief if none is produced. But Portillo is not a rule of automatic exclusion, and Figueroa declined to extend it to impeachment. Once your client voluntarily testifies, he waives the privilege against self-incrimination and can be impeached like any other witness, including with statements a judge kept out of the case-in-chief. As one defense analysis put it: a self-defense account the jury never hears is worth little, and one contradicted by an untranslated recording can cost more than silence would have, per the case breakdown.

The Figueroa appeal failed on the record, and that's the practice lesson. The defendant never showed mistranslation, never requested a continuance to translate, and never claimed cross-examination was hampered. Build these objections at the motion hearing, not on appeal.

What to do when the interview is in a language you don't read:

• Demand the recording and a certified translation early, while *Portillo* protections are strongest and before your client makes any decision to testify.
• Move pre-trial for an English transcript, and put translation-accuracy issues on the record at the hearing.
• Treat the decision to testify as inseparable from the untranslated-statement risk. Model what impeachment with that recording looks like before you advise.
• Get your own working transcript of the interview audio, plus the surrounding calls and statements, so nothing in discovery surprises you at trial.

The court's own language is worth quoting to the judge: "prudence favors translating the interview and providing a copy to defense counsel." Extending Portillo into an automatic rule is a job for the SJC or the Legislature, not the Appeals Court.

That closing gap yourself is where a grounded evidence workspace earns its keep. Evidence Copilot gives you timestamped, diarized transcripts of the interrogation recording alongside the rest of discovery, so you can compare the detective's testimony against the actual audio, and the same workflow extends to 911 call transcript analysis and jail call transcription when the case turns on what your client said, in any language, before anyone else translates it for the jury.

Building the Record at the Motion Hearing, Not on Appeal

You lose the appeal because you didn’t build the record at the motion hearing.

In Commonwealth v. Figueroa, the defendant’s conviction for assault and battery by means of a dangerous weapon was affirmed because the trial record lacked preserved objections. The appellate court noted that the defense failed to show mistranslation, request a continuance for a certified translation, or claim that cross-examination was hampered.

This tactical failure stems from a misunderstanding of Commonwealth v. Portillo. While Portillo allows judges to require English transcripts before trial, it is not a rule of automatic exclusion. The Figueroa court clarified that when a defendant voluntarily testifies, they waive privilege and can be impeached with prior statements, even if those statements were excluded from the case-in-chief due to language barriers.

To protect your client, you must establish three specific elements pre-trial. First, you must lodge a translation-accuracy objection, arguing that the interpreter’s rendition is flawed. Second, you must file a formal continuance request to obtain a certified translation from a qualified expert. Third, you must make a specific claim that the lack of a verified transcript hampered your ability to cross-examine the interrogating officer effectively.

The Figueroa decision underscores that prudence favors translating the interview and providing a copy to defense counsel. However, the judge retains discretion. If you want to preserve the issue for appeal, you cannot rely on the judge’s implicit understanding. You must explicitly state on the record that the absence of a certified translation violates due process.

This process is critical under Commonwealth v. AdonSoto, which emphasizes enhanced protections for non-English-speaking defendants. That case highlights the importance of recording interpreter-conducted interviews and ensuring the defendant understands their rights. If the police use an unrecorded, informal interpreter, you have a stronger basis for a suppression motion.

Demanding the recording and a certified translation early is your best strategy. The Figueroa court itself noted that translating the interview for defense counsel is the prudent course. By demanding this upfront, you force the Commonwealth to either provide a reliable translation or risk having the entire interrogation excluded.

Reviewing these recordings requires precise attention to detail. When you are dealing with foreign-language audio, you need a tool that can handle complex linguistic nuances and provide timestamped citations. Evidence Copilot allows you to upload the interrogation footage and generate a diarized transcript that you can verify against the police report.

This workflow helps you identify inconsistencies in the officer’s statements. You can compare the translated audio against the written police report to spot discrepancies. These inconsistencies are often the key to a successful motion to suppress.

For other types of evidence, such as initial reports, you might need to analyze the 911 call transcript analysis to see if the dispatcher’s description matches the officer’s later account. Similarly, if your client made statements from a jail cell, reviewing the jail call transcription can reveal whether the officer made promises or threats during the interview.

Building the record is about creating a paper trail. Every objection, every request, and every claim must be documented. When you get to the motion hearing, you want the judge to see that you did everything possible to ensure a fair trial. You want them to see that you asked for the translation, you challenged the accuracy, and you explained why the lack of a certified transcript prejudiced your defense.

If you skip these steps, you are gambling with your client’s freedom. The Figueroa court made it clear that the appellate court will not fix a broken record for you. You have to build it yourself, piece by piece, in the trial court.

The decision to have your client testify is also part of this equation. Testifying opens the door to impeachment. If you know that the police have an untranslated recording of a prior statement, you must weigh the risk of that statement being used against your client. A self-defense account the jury never hears is worth little. One contradicted by an untranslated recording can cost more than silence would have.

Use the tools available to you to review the evidence thoroughly. But remember, the tools are only as good as the strategy you apply to them. The law requires you to act, not just observe.

Start by demanding the certified translation. Then, build your objections into the record. Finally, prepare your client for the reality that their testimony may trigger the use of prior statements. This is how you protect your client and preserve your appeal rights.

Using AI Transcription to Close the Review Gap Without Losing the Record

Every police interrogation transcript you need but never get to review is a cross-examination question you never ask. With public defenders carrying caseloads 3 to 10 times the recommended level, the review gap isn't a workflow problem anymore. It's a due-process problem, and it's why AI transcription has moved from novelty to standard defense practice.

The reported time savings are substantial. Harris County Chief Public Defender Alex Bunin reports roughly a 50% reduction in review time when his office uses AI transcription for video evidence, according to a profile of AI in public defense. The numbers get more dramatic at the case level: one Nevada investigator reviewed 21 bodycam videos totaling about 19 hours in just 2 hours, and cut a 157-video case from roughly 140 review hours down to 12, per Thomson Reuters Institute coverage.

Miranda-warning flagging is the feature that matters most for interrogation review. Modern tools automatically surface Miranda warnings, sobriety tests, and arrest moments in the transcript, so you can jump straight to the custodial-interrogation threshold instead of scrubbing through hours of footage hoping to catch the warning. Some platforms go further, letting you ask free-form questions of the discovery itself, like whether any implicit promises were made to your client.

Before you trust any of this, hold the accuracy claims up to the light. Every benchmark in circulation is vendor-sourced: Rev cites its own 96% AI accuracy figure and a Reduct benchmark putting JusticeText at 91.54%, and none of these numbers has been independently verified. A 91.54% accuracy rate on a police interrogation transcript means roughly one wrong word in eleven. For suppression work, where the exact wording of a promise or a waiver can decide the motion, that error rate is disqualifying without human verification. As criminal and DUI lawyer Mike Kruse puts it, technology "does not eliminate the necessity of critical thinking and judgment at the preparation stage."

Then there's the data-handling question TechCrunch's AI editor flagged when covering JusticeText: you're uploading your client's most sensitive evidence to a third party. Before any upload, vet the following:

• Encryption of evidence at rest and in transit, and who can access it.
• Whether your uploads can be used to train the vendor's models, and whether you can opt out.
• Redaction capability, so you can turn over discovery without exposing bystander faces, PII, or privileged material.
• Whether AI output is marked as pending human review, and whether every answer cites the exact source moment so you can verify against the original recording.

The workflow that holds up under ethical scrutiny treats the AI transcript as a finding aid, never the record. Use it to locate the Miranda warning, flag the inconsistency between the offense report and the bodycam, and build your timeline. Then verify every quote you intend to use against the original video. Tools built for defense work, like Evidence Copilot, keep originals immutable and mark every AI output for human review precisely because the transcript you cite in a motion has to be the one you checked yourself. The same standard applies across your evidence stack, whether you're working with a 911 call transcript analysis workflow or jail call transcription for a client facing recorded calls from custody.

A Grounded Review Workflow You Can Defend to the Court

You can't defend a transcript you never built a record around. In Commonwealth v. Figueroa, the Appeals Court affirmed a conviction partly because the defense never showed mistranslation, never requested a continuance to translate, and never claimed cross-examination suffered, as this defense analysis of the Figueroa ruling lays out. Under Commonwealth v. Portillo, a judge may require an English transcript before trial and may exclude foreign-language statements from the case-in-chief, but that is discretionary, not automatic. So the workflow starts before any AI tool touches a file.

Request the interrogation recording, the 911 audio, and jail calls at arraignment, and put the translation demand in writing. The Figueroa court itself noted that prudence favors translating an interview and giving defense counsel a copy. Build your objection record at the motion hearing, because appellate courts won't rescue an argument you never preserved.

Public defenders carry caseloads three to ten times the recommended amount, according to TechCrunch's reporting on AI transcription in public defense. You cannot manually review that volume at scale, but you also can't trust a black-box summary. Run timestamped, diarized transcripts on interrogation footage, 911 calls, and jail calls so every quote maps to an exact moment.

• Demand the recording and a certified translation in writing at arraignment.
• Transcribe interrogation footage, 911 audio, and jail calls with speaker labels and word-level timestamps.
• Cite every AI-generated answer to the exact source moment, and mark it pending human review.
• Preserve translation-accuracy objections, continuance requests, and cross-examination claims at the motion hearing.

Reported time savings from AI review run near 50%, per a Stand Together profile of Harris County's chief public defender, but accuracy benchmarks are vendor-sourced and vary widely (91.54% to 99% depending on who measured it). The rule that keeps you defensible: every AI answer cites the exact source moment it came from, and every output is marked for human review. Evidence Copilot builds to that standard, with one-click speaker renaming that propagates across segments and immutable originals.

On cost, JusticeText runs roughly $1,200 per attorney per year, Rev charges $0.25 per minute for AI and $1.99 per minute for human transcription, and Evidence Copilot publishes Solo at $39 per user per month and PD Office at $750 per attorney per year. If you want to see the cited workflow on your own footage, book a 20-minute walkthrough at evidencecopilot.com/book.