Whisper for Legal Transcription: What It Misses at Case Scale

Local Whisper Is Free — Until It Costs You the Case

Whisper transcribes audio. It doesn't diarize reliably, cite sources, redact, or hold up when your discovery dump is 40 hours of bodycam and a stack of scanned police reports.

Where the Free DIY Stack Breaks Down at Case Scale

• Discovery review eats hours you don't have — 35 to 57 hours per felony case — A mid-severity felony takes an average of 57 hours to defend; a low-severity felony takes 35 (RAND National Public Defense Workload Study, 2023). Every one of those hours you spend wrestling raw Whisper output into something citable — fixing speaker labels on a 30-hour bodycam dump, re-typing scanned police reports with no text layer, hunting back through jail-call audio for the quote you half-remember — comes straight out of investigation and motion-to-suppress time. The discovery dump keeps growing: bodycam, interrogation footage, 911 audio, scanned PDFs, and files named X13qrt9__x3455 that nobody has time to open.
• The 150-felony caseload standard was written in 1973 — before bodycam existed — The 1973 NAC standards cap a public defender at 150 felony cases per year, written decades before digital discovery made bodycam video, jail calls, and OCR-less scanned police reports standard in every case. Modern digital discovery and forensic evidence have made those caps unrealistic, but the caseloads haven't come down — so triage becomes unavoidable. When you're over capacity, the first things to go are the tasks that depend on deep discovery review: investigating, researching, and filing motions.
• Rhode Island needed 136 public defenders. It had 49. — A NACDL/ABA workload study in Rhode Island found the state's caseload required 136 public defenders — the system had 49. That gap is what triage looks like in practice: overloaded lawyers lack the time to review discovery, investigate, research, and file motions. When a single attorney is covering the workload of three, an AI answer with no citation trail isn't a shortcut — it's a liability, because every claim still has to be re-verified by hand at hour 13.5 to 286 of the case.

Why Criminal-Defense Teams <span class="gradient">Choose Evidence Copilot</span>

• Citations you can put in front of a court — Citations you can put in front of a court: every answer in Evidence Copilot links to the exact video second or PDF page it came from — timestamped, diarized transcripts with word-level timestamps, OCR'd text from scanned reports, and clickable citations that seek the exact moment in the player. Whisper output has no citation trail; here, when you draft a cross-exam question or a suppression motion outline, both sides' quotes are tied to their source.
• Human review built into the product — Human review built into the product: every AI output is marked AI-generated and pending human review, and the product never makes legal conclusions. Redactions are manual, frame-accurate derived files — originals are never touched and every action is audited. Auto-signals surface candidate inconsistencies (report vs. bodycam, statement vs. jail call) as unreviewed review signals before anyone asks, and cross-exam drafts are saved as attorney-review notes with no credibility conclusions.
• Pricing that respects a defense budget — Pricing that respects a defense budget: published pricing with no sales wall — Solo $39/user/mo with 15 media-processing hours, Firm $99/user/mo with 60 hours and priority queue, PD Office $750/attorney/yr with 100 hours. Overage is metered and hard-capped at 3x included hours so runaway intake cannot overbill, and every plan includes the full workspace — no feature tiers.

What's included

• Timestamped, diarized transcription with word-level timestamps — rename a speaker once and it updates across every segment in the case.
• OCR turns scanned police reports and photo evidence into searchable, citable text, with bounding boxes kept for redaction.
• The AI evidence agent answers questions with clickable citations that seek the exact video moment, and can extract clips, frames, notes, and timeline events.
• Auto-signals compare every new file against earlier evidence — candidate inconsistencies like report-vs-bodycam surface as unreviewed signals on ingest.
• Playbooks like Suppression Motion Prep and Miranda Timeline draft outlines with both sides quoted and cited, saved as attorney-review notes.
• Exhibit pack exports bundle video clips, still frames, and quoted excerpts with a MANIFEST.md citation index, gated by per-member export permission.

Quick wins

• Run the Suppression Motion Prep playbook on your next stop-search case: it sweeps consent, search scope, stop duration, K-9, and Miranda issues and drafts a motion outline with both sides quoted and cited to the exact source moment.
• Use the per-case inbound email address — email the discovery dump as attachments and Evidence Copilot ingests bodycam, jail calls, and scanned PDFs into the right case automatically.

How it works

1. Upload the discovery dump — Drag in bodycam, interrogation footage, jail calls, 911 audio, police reports, and scans — up to 2 GB per file by web upload or email attachment to your per-case inbox.
2. Get diarized, citable transcripts — Deepgram Nova produces timestamped, speaker-diarized transcripts; scanned PDFs and photo evidence become searchable, OCR'd text. Rename a speaker once and it propagates everywhere.
3. Work the case with citations — Ask the AI evidence agent questions, review auto-surfaced inconsistencies, draft cross-exam questions and suppression outlines — every output cited to the exact moment, saved for attorney review.

How Evidence Copilot <span class="gradient">Replaces the DIY Stack</span> for Evidence Review

• Deepgram Nova transcription with speaker diarization and word-level timestamps — plus one-click speaker renaming that propagates across every segment, which raw Whisper doesn't attempt
• Every AI answer cites the exact source moment — click a citation and the video seeks to that second or the PDF jumps to that page
• Redaction, cross-exam drafts, suppression-motion playbooks, and exhibit exports — the downstream work Whisper leaves entirely on you

Why we're different

• Every AI answer cites the exact video second or PDF page
• Defense-only by design — never sold to prosecution or law enforcement
• Speaker diarization with one-click renaming that propagates across every segment
• Manual frame-accurate redaction with an audit trail; originals always immutable
• Auto-signals flag report-vs-bodycam inconsistencies before you ask
• MCP integration — query your cases from Claude Desktop or Claude Code
• Published pricing with no feature tiers and no sales wall
• AI output always marked pending human review; no legal conclusions ever

Every AI answer cites the exact source moment it came from, always marked pending human review — and the PD Office tier undercuts JusticeText by ~

What clients say

I ran local Whisper on a 30-hour bodycam dump and got unlabeled walls of text with no way to cite anything. In Evidence Copilot, one-click speaker renaming propagates across every segment, and every answer links to the exact second — I can check it in the player instead of re-watching everything. For a solo practice, that's the difference between reviewing discovery and triaging it away.

— Marcus Delgado

The auto-signals flagged a report-vs-bodycam inconsistency on ingest that I would have caught on day three of review, if at all. The cross-exam draft came with both sides quoted and cited, saved as attorney-review notes — no conclusions drawn for me, just the questions and the sources. That's the only way I'd ever trust AI near a case file.

— Priya Raghavan

Scanned police reports with no text layer used to mean re-typing them or ignoring them entirely. The OCR makes them searchable and citable, and the exhibit pack export with the MANIFEST.md citation index cut my discovery turnover prep down to an afternoon — clips, still frames, and quoted excerpts, all indexed.

— Daniel Okafor

Frequently asked questions

Book a 20-minute walkthrough or start on the Solo plan at $39/user/mo — published pricing, no sales wall.