The Real Cost of Transcribing Discovery: It's Not the Per-Minute Rate

You get a 40-hour bodycam dump, three hours of jail calls, and a 911 recording — all due before the preliminary hearing. The question "how much does transcription cost?" sounds simple. It isn't, because the headline per-minute rate is the wrong number to shop on.

The market spans a roughly 250x gap between the cheapest and most expensive options, according to one legal transcription pricing guide. Human legal transcription runs $1.50–$5.00 per audio minute — that's $90–$300 per audio hour. AI transcription typically costs $0.10–$0.25 per minute, or $6–$15 per audio hour. Certified court reporters charge $4.40–$8.70+ per page, with federal maximum rates set by the Judicial Conference as of October 2024.

Run those rates against a 40-hour bodycam dump and the math gets ugly fast. At human rates, that single dump implies $3,600–$12,000 in transcription before you touch the jail calls. At AI rates, it's $240–$600. But the sticker rate is only the beginning.

The surcharges are where vendors actually make their margin. Common add-ons include:

• Rush or expedite fees — typically a 50–100% premium on the base rate
• Multi-speaker fees — roughly $0.25/min extra for recordings with 3+ speakers
• Poor-audio surcharges — $0.50–$2.00/min on crosstalk, wind, or distant mics
• Timestamping add-ons — another $0.25–$0.50/min, often essential for impeachment work
• Minimum order fees — $50–$250, which stings when you only need one 911 call transcribed

Notice what that list means for discovery work specifically. Bodycam footage is multi-speaker, jail-call audio is poor quality, and every transcript you cite needs timestamps. The recordings that defenders actually handle trigger nearly every surcharge on the menu. A $1.50/min quote can quietly become $3.00/min on real evidence.

There's also the labor cost on the back end. Analysis of free-tier AI transcripts found correcting a 60-minute recording took 150–240 minutes of editing — $50–$80 of attorney or clerk time at $20/hour. That "cheap" transcript isn't cheap if someone on your team rebuilds it manually.

One caveat on all these figures: most pricing data comes from vendor blogs with commercial interests, not independent market surveys. The most verifiable data points are the federal court reporter rate schedule and a peer-reviewed bodycam study. Treat every per-minute number — including the ones above — as directional, not gospel.

The practical takeaway for defense teams: stop shopping per-minute rates and start comparing total cost of ownership. That's the logic behind Evidence Copilot's published pricing — Solo at $39/user/month with 15 media-processing hours included, Firm at $99 with 60 hours, and a PD Office tier at $750/attorney/year with 100 hours, with overage hard-capped at 3x included hours so a big intake dump can't balloon the bill. A flat plan that bundles diarization, timestamps, and OCR beats per-minute metering once you're processing more than a handful of recordings a month — which, in criminal defense, is every case.

Per-Minute vs. Flat Plans: Doing the Math on a Real Case

Here's what those rates mean when the county attorney's office drops a 40-hour discovery dump — bodycam, interrogation footage, jail calls — on your desk two weeks before the preliminary hearing.

At human transcription rates of $1.50–$5.00 per audio minute, which works out to $90–$300 per audio hour according to a legal transcription pricing guide, that single dump costs $3,600 to $12,000. At AI rates of $0.10–$0.25 per minute, the same 40 hours runs $240 to $600, per AI-vs-human transcription comparisons. Same evidence, a 15x difference in price — and that gap is why the pricing model you pick matters more than the headline rate.

Transcription vendors bill three ways, and each fits a different workload:

• Per audio minute — pay only for what you send; sensible for a one-off deposition or a single 911 call you need certified.
• Per word — common for document-heavy legal discovery work, where the deliverable is a page count, not a recording.
• Flat per-seat subscription — a fixed monthly cost with included hours, built for recurring volume.

The research finding that matters for defense work is blunt: once you're transcribing more than a handful of calls a month, a flat plan beats per-minute metering, often by a wide margin. A defender handling even two cases with media evidence blows past "a handful" immediately. And per-minute pricing on discovery volume invites the surcharge trap — timestamping fees, multi-speaker fees for 3+ voices, poor-audio add-ons of $0.50–$2.00 per minute. Jail-call audio checks all three boxes.

This is where flat, published pricing changes the math. Evidence Copilot's plans are structured around included processing hours rather than metered audio minutes: Solo at $39/user/month with 15 hours, Firm at $99/user/month with 60 hours, and a PD Office tier at $750/attorney/year with 100 hours. Overage is metered but hard-capped at 3x the included hours, so a surprise 80-hour intake can't turn into an unbounded invoice. Transcription with speaker diarization, OCR on scanned reports, timestamped citations, and review tooling are bundled — you're not paying separately for each feature a per-minute vendor lists as an add-on.

Two honest caveats. First, subscription value depends on utilization — a plan you underuse costs more per hour than metering, which is why solo defenders with light media volume should start small. Second, cheap AI transcription isn't a substitute for certified transcripts where admissibility requires them; the smart pattern vendors and researchers both describe is AI for the searchable working draft, human or certified transcription only for the passages that end up in front of a court.

Run your own numbers: take last month's discovery hours, price them at $90–$300 per audio hour versus a flat plan, and the right model usually picks itself.

Why Cheap AI Transcripts Cost More Than They Look: The Accuracy Gap

A $0.10-per-minute AI transcript sounds like a bargain until you sit down to fix it. One industry analysis found that error correction on free-tier AI transcripts runs 150–240 minutes of editing per hour of audio — $50–$80 of your time at a $20/hour billing rate, spent cleaning up a transcript that cost almost nothing upfront (BrassTranscripts).

The problem isn't just typos. A peer-reviewed Northeastern study of 176 bodycam videos (~23 hours, 73 incidents) found that Rev.ai combined statements from different speakers, identified an average of 3.3 participants per transcript versus 4.8 in human-edited versions, and dropped roughly 25,000 words across the sample. Kansas City PD stopped using Rev.ai on bodycam footage entirely. As EPIC's Calli Schroeder put it, "an AI transcription or summary may be 98% accurate, but that 2% could completely change who is at fault."

For defenders, that 2% is often the impeachment gold: who said what, in what order, on which camera. Raw word accuracy doesn't capture a merged speaker attribution or a missing exchange between an officer and a witness.

The practical answer isn't "never use AI" — it's the hybrid workflow. AI handles discovery drafts; human or certified transcription is reserved for filings where certification matters. Because research shows 80–90% of legal transcription work doesn't require court certification, this approach cuts transcription costs 60–80% while keeping admissibility intact.

The workflow only works, though, if the AI layer is built for review rather than blind trust:

• Speaker diarization with word-level timestamps, so you can see exactly who is attributed to each line and fix misattributions fast.
• One-click speaker renaming that propagates across every segment, instead of editing "Speaker 2" forty times.
• Every AI output marked AI-generated and pending human review — never presented as the record itself.
• Citations that link each answer to the exact video moment, so you verify against the original, not the transcript.

That last point matters because courts have ruled AI transcripts inadmissible without human verification. The transcript is a working tool; the recording is the evidence. Any review setup that lets the two blur is where cheap AI starts costing real money — in edit time, in missed inconsistencies, in a suppression hearing where the wrong speaker is on the record.

This is the design logic behind Evidence Copilot: Deepgram Nova diarization with word-level timestamps, one-click speaker renaming, and every AI answer citing the exact source moment it came from, with originals kept immutable. The AI does the heavy lifting on a 40-hour bodycam dump; the human decides what the details mean — and the pricing reflects that division of labor, with processing hours included in flat plans rather than metered per audio minute.

What to Ask Before You Buy: A Practical Cost Checklist

Before you sign anything, run the vendor through five questions. The transcription pricing research is blunt on this point: the headline per-minute price is the wrong number to shop on, because surcharges, caps, and correction labor dominate what you actually pay. Here's the checklist we'd hand a defense team.

1. Calculate your monthly audio hours — not the sticker rate. Convert everything into per-audio-hour figures you recognize. Human legal transcription at $1.50–$5.00/min works out to $90–$300 per audio hour; AI at $0.10–$0.25/min is $6–$15 per audio hour. Against a 40-hour bodycam dump, those rates imply very different invoices. Then compare subscription value honestly: a vendor's monthly plan only "wins" if you use all included hours — using half of them can double your real per-hour cost.

2. Ask about surcharges in writing. Legal-specific add-ons are where budgets break: multi-speaker fees (+$0.25/min for 3+ speakers), poor-audio surcharges (+$0.50–$2.00/min), timestamping (+$0.25–$0.50/min), verbatim surcharges, expedite premiums (50–100%), and minimum order fees ($50–$250). Jail calls and interrogation footage hit three of those triggers on day one — multiple speakers, hard audio, timestamps you need for cross-exam.

3. Check overage and storage caps. Runaway intake — the discovery dump that arrives mid-trial — is exactly when metered pricing overbills you. Ask what happens when you exceed included hours and whether the vendor caps overage. Evidence Copilot's published pricing, for example, hard-caps overage at 3x included hours so a heavy month can't spiral; Solo includes 15 processing hours at $39/user/mo, Firm 60 hours at $99, and the PD Office tier is $750/attorney/yr with 100 hours — with the full workspace on every plan, no feature tiers.

4. Confirm diarization quality and citability. The most credible data in this whole market is academic: a peer-reviewed Northeastern study of 176 bodycam videos found AI misattributed speakers, identified 3.3 participants per transcript versus 4.8 in human-edited versions, and dropped roughly 25,000 words across the sample. So don't ask "does it transcribe" — ask "does it separate speakers correctly, and can I cite the exact video second?" A transcript you can't pin to a moment isn't useful for impeachment or a suppression hearing.

5. Verify AI outputs are marked for human review. Courts have ruled AI transcripts inadmissible without human verification, and as Calli Schroeder of EPIC put it, a summary may be 98% accurate but "that 2% could completely change who is at fault." Every AI output in Evidence Copilot is marked AI-generated and pending human review, with citations into the exact source moment — and the originals stay immutable.

One last anchor: a Princeton study of public defense found cost is a documented barrier to AI adoption, while defenders see the greatest promise in evidence investigation — transcribing and synthesizing large volumes of digital records. Transcription is the highest-value use case, which is why it's worth getting the pricing model right.

If the checklist surfaces a fit, compare against Evidence Copilot's published pricing at evidencecopilot.com/pricing, book a 20-minute walkthrough at evidencecopilot.com/book, or start on the Solo plan at evidencecopilot.com/signup. No pressure either way — the math is the math.