Legal transcription for $2 per hour
Depositions, hearings, client interviews, recorded statements. Speaker labels separate every voice in the room — useful for multi-party proceedings where attribution matters. AI-assisted, human-reviewable, with EU data processing for confidentiality.
Speaker labels for depositions
A 5-party deposition with attorneys, witness, and court reporter produces a single mixed audio track. Our diarisation labels each voice — the editor lets you rename to Counsel, Witness, Court Reporter, etc.
EU data processing
Audio is processed in EU data centres and deleted from our servers 90 days after your last sign-in. We don't train models on your audio. Zero-retention agreement with our transcription subprocessor.
$2/hour, no hidden fees
A 4-hour deposition costs $8. A 1-hour client interview costs $2. No per-page fees, no expedited surcharges, no rush rates.
Human-reviewable, not authoritative
AI transcription is fast and accurate (90–97% on clean audio) but not court-certified on its own. Use TranscribeCat as a reviewer's starting point, then human-verify before official use.
From recorded proceeding to transcript in 3 steps
Upload the recording
Drop a WAV, MP3, M4A, or MP4 of the proceeding. Up to 500 MB / 10 hours per file. For longer or larger files, FLAC compression or splitting on a natural break works fine.
We transcribe and label speakers
Our pipeline detects the language (90+ supported, useful for client interviews not in English), separates speakers, and generates a verbatim transcript with timestamps.
Export and review
Download as Word document with speaker labels and timestamps. Use the in-app editor to rename speakers, mark sections, and verify against the audio playback before finalising.
Depositions specifically — what makes them hard, and what helps
Depositions have an unusual audio signature compared to other professional recordings: multiple voices in one room (often 4–6 people), strict turn-taking that breaks down during objections, technical legal vocabulary that varies wildly by practice area, and frequent references to exhibits by number that the transcript needs to preserve faithfully.
What helps AI transcription on this kind of audio:
- Lapel mics on each speaker: changes the math entirely. Cross-talk is reduced, individual voice levels are stable, diarisation accuracy goes up. If you regularly transcribe depositions, the investment in 4 lavalier mics + a multitrack recorder pays back fast.
- A single boundary mic on the conference table: second-best. Better than a phone or laptop mic. The geometry of who's sitting where determines speaker clarity.
- A clean room, not a courtroom hallway: HVAC, foot traffic, and hallway echo all hurt accuracy. A small conference room with closed doors and carpet is acoustically ideal.
- A pause-and-acknowledge protocol: when speakers cut each other off, transcription always struggles. Even a small protocol — "Counsel, please complete your sentence" — produces meaningfully cleaner transcripts.
Realistic accuracy expectations on a typical deposition: 93–96%with lapel mics, 88–93% with a single conference mic, 80–87% with phone-on-table audio. AI transcription is a strongstarting pointfor a human reviewer to clean up — it's not a replacement for a certified court reporter where one is required.
Confidentiality, chain of custody, and what TranscribeCat is (and isn't)
Legal work has stricter data handling than most industries, and we want to be clear about what TranscribeCat is suited for:
- Internal client interviews, witness prep, fact-finding recordings: well-suited. Upload, transcribe, review. Audio deleted 90 days after your last sign-in.
- Depositions and hearings (informal use): well-suited as a working reviewer's draft. The transcript is for your team's review, not for court filing. A certified court reporter remains the authoritative source for the official record.
- Recorded statements, investigation interviews: well-suited. 90+ language support is useful when interviewees aren't native English speakers.
- Privileged communications: review your firm's policy on third-party processors first. Our subprocessors are listed at /trust. We process in EU data centres and don't train models on your data, but each firm has its own threshold for what can leave the firm's direct control.
- Court-certified transcripts (official record): not suited. AI transcription, regardless of accuracy, is not currently certified for use as the official court record. Use a certified court reporter for that.
What we don't do: claim certifications we don't have. TranscribeCat is not HIPAA-certified, not court-certified, and not a SOC2 Type II vendor. We are SOC2-aligned in practice and transparent about our subprocessors at /trust.
What legal transcription costs
$2 per hour, regardless of complexity or speaker count:
$2
1-hour client interview
$8
4-hour deposition
$16
8-hour hearing day
Compare: human legal transcription is typically $2.50–$5 per audio minute, or $150–$300 per hour. AI transcription gets you 90% of the way there for ~1% of the cost — your reviewer’s time is the variable.
Frequently asked questions
Can I use TranscribeCat for court-certified transcripts?+
No — AI transcription, however accurate, is not currently certified for use as the official court record. A certified court reporter remains the authoritative source. TranscribeCat is best used as a working reviewer's draft, witness prep transcript, or internal interview record where certification isn't required.
Is the audio kept confidential?+
Audio is processed in EU data centres and deleted from our servers 90 days after your last sign-in. We don't train models on customer audio. Our transcription subprocessor (OpenAI) operates under a zero-retention agreement. Full subprocessor list at /trust. For privileged communications, verify with your firm's policy that non-BAA processors are permissible.
How accurate is transcription on legal vocabulary?+
Whisper-class models handle most legal terminology well — they've been trained on broad English corpora that include legal text. Specialized practice-area vocabulary (patent law jargon, niche statutes) may produce occasional misspellings that a human reviewer can correct. Realistic accuracy on clean deposition audio: 93–96%.
Can I export with speaker labels for review?+
Yes. Word document export includes speaker labels and timestamps. Rename speakers in the editor first (Speaker 1 → Counsel, Speaker 2 → Witness) so the export is review-ready.
What about non-English client interviews?+
90+ languages supported with auto-detection. The transcript stays in the source language — we don't auto-translate. For evidence purposes, having a verbatim transcript in the source language is usually preferred (translation is a separate, attested step).
How long does a 4-hour deposition take to transcribe?+
Typically 12–20 minutes from upload completion to transcript ready. The page updates automatically; you can leave the tab and come back. For very long files (8+ hour hearing days), splitting on a natural break and transcribing in parallel is the fastest path.
Can I share the transcript with the rest of my team?+
Yes — TranscribeCat supports organisations with shared billing. One firm card pays for all uploads by all team members. No per-seat fees. Set up at /org/settings.
Related legal and confidentiality resources
Transcribe your next deposition or interview
Speaker-labeled, EU-processed, $2 per hour. Your reviewer's starting point, not the official record.
Start Transcribing ($2/hr)Free to sign up · Pay only when you transcribe