VocalFuse is a Fuse Intelligence product.

LEGAL DEPOSITION TRANSCRIPTION

AI Deposition Transcription — Working Copies That Never Leave Your Machine

A certified deposition transcript runs $1,000–$2,500 all-in. Your litigation team needs a searchable working copy of every deposition — and privileged audio uploaded to a cloud service is a disclosure surface United States v. Heppner (S.D.N.Y. 2026) put on notice. VocalFuse transcribes with local Whisper on your Windows PC: no upload, no per-minute meter, no vendor copy to subpoena — $5/mo or $79 one-time.

What a deposition transcript actually costs in 2026

Deposition transcription is one of the few legal services where three pricing models coexist — per page of transcript, per minute of audio, and per hour of professional time — which makes quotes nearly impossible to compare without converting. Here is the landscape.

Cost component Typical 2026 rate Notes
Appearance fee $300–$700 per session Charged just for showing up; two-hour minimums common; real-time reporters $600–$1,000+ (LegalClarity)
Original transcript (per page) $4.50–$7.00 private market Judicial Conference federal caps: $4.40 ordinary → $8.70 two-hour turnaround (in effect since Oct 2024)
Copy charges (per party) $1.10–$1.45 federal cap Private-market copy rates run higher; every ordering party pays separately
Expedited delivery +30–50% surcharge Next-day "daily copy" can nearly double the standard rate ($4.40 → $7.30 federally)
Real-time feed $600–$1,000+ per session Requires specially certified reporter; premium hard to justify for routine fact witnesses
All-in, typical deposition $1,000–$2,500 Half-day ≈ 100–150 pages; full-day can exceed 300 pages — per-page fees scale linearly
AI working copy (cloud) $0.25–$0.30 per audio minute Rev AI $0.25/min, Sonix $10/audio-hr; two-hour deposition ≈ $30 — minutes, not days
AI working copy (local) $0 marginal Local Whisper after one-time install: same audio, again and again, for free — and nothing leaves the machine

Sources: LegalClarity deposition-cost explainer (Apr 2026), United States Courts Federal Court Reporting Program rate schedule, TranscribeLegal 2026 pricing guide, vendor pricing pages (Rev, Sonix). Rates vary by metro and agency; verify current quotes before budgeting.

The two-track workflow law firms actually use

A deposition generates two different transcription needs. Treating them as one problem overpays for both (BrassTranscripts, Aug 2026).

Track 1: the certified record

A certified court reporter produces the official transcript for anything that will be filed, entered into evidence, or used at trial. FRCP 30(b)(3) requires the testimony to be recorded by the officer or someone acting under the officer's direction — an AI draft cannot serve this role. Costs $500–$2,000+, delivered in days.

Track 2: the working copy

Everything else your team does with testimony — searching it, summarizing it, checking it against prior statements, building the inconsistency file, prepping witnesses — needs a fast, cheap draft. AI transcription returns it in minutes at roughly one percent of the certified price. This is where VocalFuse lives.

Where the working copy earns its keep:

  • Same-day review. The AI draft arrives while the deposition is fresh — inconsistencies get flagged within 48 hours, not after the certified transcript lands next week.
  • Every deposition gets transcribed. At $750+ per certified original, firms ration. At $0 marginal, every fact-witness deposition, witness interview, and intake call gets a searchable transcript.
  • The 2026 summary layer. AI deposition-summary tools now process a 100-page transcript in about three minutes with page-line citations (thelawgpt, LuMay benchmarks) — but they all need a transcript first. The working copy feeds them.
  • Cost-recovery hygiene. Copy charges get taxed as costs under 28 U.S.C. § 1920(2); working copies for convenience often do not. Keeping the cheap layer cheap keeps the bill defensible.

The privilege problem with cloud deposition transcription

In United States v. Heppner (S.D.N.Y., decided February 2026), Judge Jed Rakoff held that documents created using a consumer AI tool were not protected by attorney-client privilege — the tool's terms of service permitted data retention and third-party disclosure, which destroyed the confidentiality privilege requires (ConvertAudioToText legal analysis; Basil AI privilege ranking). The reasoning tracks the third-party doctrine: voluntarily sharing content with a service provider weakens the confidentiality expectation.

  • The disclosure happens at upload. Privilege attaches at the moment of communication and belongs to the client. Once audio leaves your machine to a vendor whose terms you have not fully vetted, the waiver question is live — and unlike most legal errors, there is no retrospective fix (Norton Rose Fulbright's UK-equivalent analysis: waived privilege cannot be recovered).
  • Subpoena surface. Stored audio on a vendor's infrastructure is subject to compelled production — the same § 2703 mechanics that reach journalists' cloud tape reach deposition audio. Vendor breach disclosure adds a second exposure.
  • The vendor has the audio. Even NDA-bound human transcriptionists listen to every minute. For witness-prep recordings, strategy sessions, and attorney work product, that is a third party in the room.
  • Bar guidance is converging. ABA Formal Opinion 512 guidance and NYC Bar Formal Opinion 2025-6 both push lawyers to understand a tool's data flow before feeding it privileged material. Local processing is the only architecture where the question never arises.

Enterprise tools under DPAs occupy a different posture than consumer apps — but the cleanest answer to "could this upload waive privilege" is an architecture where nothing is uploaded.

Cloud vs local: where your deposition audio actually lives

Question Cloud AI transcription (Rev, Sonix, Otter) Local transcription (VocalFuse)
Who holds the audio? The vendor's storage — a subpoena surface in litigation against your client or the vendor Your PC only. No provider copy exists to compel
Heppner disclosure risk Live — uploading to a third party is the act the court examined None — no disclosure occurs because nothing is transmitted
Two-hour deposition cost $30 (Rev AI $0.25/min) to $10 (Sonix) per audio hour, metered forever $0 marginal per file — flat $5/mo or $79 one-time
Offline / courtroom use Requires upload; connectivity dependent Runs fully offline — works in interview rooms with no Wi-Fi
Vendor training on your audio Varies by provider terms; opt-outs exist but require vigilance Impossible — no server receives the audio
Certified transcript? No — cloud AI is a draft, same as local No — working copy only; certified record still comes from a reporter

The local deposition workflow in four steps

1. Record the session

Capture the deposition audio locally — a recorder, a video call's local audio track, or your own microphone. For remote depositions, capture the audio feed on your machine instead of trusting the platform's cloud recording.

2. Transcribe on-device

Drop the file into VocalFuse. A local Whisper model transcribes on your Windows PC — minutes per hour of audio, with speaker structure captured for the attorney/deponent split. No upload step exists to audit.

3. Verify in the editor

Read the transcript on screen against the audio for names, terms of art, and crosstalk. This review is your duty of competence over AI-drafted work product — treat it exactly like verifying a junior associate's digest.

4. Export under your rules

Export the working copy into your case file with your own access controls. The raw audio and the transcript never existed anywhere but your machine — your retention schedule is the only one that applies.

Where VocalFuse fits — and where it does not

  • Working copies, prep sessions, intake calls. Everything your litigation team uses internally — searchable drafts in minutes, zero marginal cost, nothing on a vendor's server.
  • Attorney dictation and case notes. The same local engine handles dictated memoranda and hearing notes with no per-minute meter.
  • Not the certified record. Filings and evidence need a certified court reporter under FRCP 30(b)(3). We do not pretend otherwise — the two-track workflow uses both.
  • Flat pricing. $5/mo or $79 one-time. No per-minute billing that penalizes long depositions, no AI credits that expire mid-discovery. See VocalFuse features.

Related reading

Transcription for journalists

The same § 2703 subpoena mechanics applied to source protection — the closest privacy analog for privileged legal audio.

Read the journalist guide

AI notetaker privacy

Voiceprints, consent, and the 2026 litigation wave reshaping what legal teams allow.

Read the privacy guide

Otter AI alternative

Bot-free, local alternative to the most common meeting-transcription default.

Compare with Otter

Meeting minutes generator

AI-drafted minutes and action items from local transcription — the intake-call and client-meeting companion.

Generate minutes locally

Explore related AI note taking guides

AI deposition transcription — FAQ

How much does a deposition transcript cost?

A typical certified deposition transcript costs $1,000–$2,500 all-in once appearance fees and per-page charges are tallied (LegalClarity, April 2026). Private-market original-transcript rates generally run $4.50–$7.00 per page; a half-day deposition produces 100–150 pages, so per-page fees alone run $750+ at a $5.00 rate — before the $300–$700 appearance fee. Federal courts cap originals at $4.40–$8.70 per page depending on turnaround, but private agencies are not bound by those caps.

What is a deposition working copy and why do firms need one?

A working copy is the fast, cheap transcript your litigation team actually uses for search, summary, and cross-reference work — not the certified record that gets filed. Firms run the deposition audio through AI transcription to get a searchable draft in minutes instead of waiting 7–10 days for the certified original (BrassTranscripts calls this the two-track workflow). You need one per deposition; you only need the certified original for filings and evidence.

Can AI transcription of privileged audio waive attorney-client privilege?

It is a live risk. In United States v. Heppner (S.D.N.Y., Feb 2026), Judge Rakoff held that content shared with a consumer AI platform whose terms permitted retention and disclosure was not privileged — voluntary disclosure to a third party undermines the confidentiality privilege requires. Enterprise tools under DPAs are a different posture, but the safest architecture is not to disclose at all: local transcription never creates the third party in the first place. This page is general information, not legal advice.

Is VocalFuse a substitute for a certified court reporter?

No — and it should not be framed as one. The official, certified transcript for filings and evidence still comes from a court reporter under FRCP 30(b)(3), which requires the officer or a person acting under the officer's direction to record testimony. VocalFuse produces working copies for case preparation, discovery review, witness-prep notes, and intake calls — everything your team does with testimony besides entering it into evidence.

What does AI deposition transcription cost compared to per-page rates?

Cloud AI services charge per audio minute: Rev AI at $0.25/min (~$30 for a two-hour deposition), Sonix at $10/audio hour, or AI-assisted human review at $1.00–$1.50/min. A local Whisper setup has zero marginal cost per file — the same audio transcribes again and again for free after the one-time install. VocalFuse is a flat $5/mo (or $79 one-time) with no per-minute metering, so a two-hour deposition costs the same as a two-minute intake call: nothing extra.

Where does my deposition audio go with VocalFuse?

Nowhere. VocalFuse transcribes with a local Whisper model on your Windows PC — no upload, no cloud speech API, no vendor account holding your audio. There is no vendor server holding privileged material to subpoena, breach, or compel, because no copy exists outside your machine. On the optional Pro plan, text notes and summaries can sync to your account — for privileged matters, keep sync off and everything stays on the device.

Do AI transcripts handle legal terminology and multiple speakers?

Modern Whisper-class models handle legal phrasing, case names, and citations reasonably well on clean audio, and VocalFuse labels speakers (attorney, deponent, witness) in its structured notes — verify both in the on-screen editor, exactly as you would verify a junior associate's summary. Expect proper names and heavily crosstalk audio to need human correction; that review is the same duty you already owe every AI-drafted work product.

Which depositions should still get the certified transcript?

Any deposition that will be filed, entered into evidence, used at trial, or cited in a dispositive motion needs the certified original — courts scrutinize "necessarily obtained" transcripts under 28 U.S.C. § 1920(2) at costs taxation. Routine fact-witness depositions used only for preparation and discovery organization are where the working copy carries the load. Many firms do both: certified original from the reporter, AI working copy for the team.