A deposition is the most quotation-sensitive conversation most professionals will ever sit through: every word can be read back, filed, and argued over. The same is true of the witness statement a journalist takes before publication, and of the client interview a lawyer's paralegal summarizes. These recordings cannot take the ordinary upload path — sending case material to a transcription service puts third parties inside a chain that privilege rules and discovery obligations may reach. Transcribing locally is the arrangement that keeps the chain at one link: you.
The short version
- QuickSpeak transcribes on your machine — deposition audio is captured and recognized in WebAssembly in your browser; the transcript and audio are files on your disk. No upload, no account, nothing for discovery to subpoena from a vendor.
- Speakers come labeled (attorney/witness/interpreter) via on-device diarization; rename to roles and quote from the transcript, not from memory.
- Machine transcript = working draft. Verify every quotable phrase against the audio — which stays paired with the transcript locally, one click away.
- Free tier covers it — unlimited transcription, full history, .txt export. Pro ($12/mo or $99/yr, 14-day trial) adds Markdown/JSON export, 500MB file upload for recordings made elsewhere, and speaker profiles.
Who needs this workflow
- Attorneys and paralegals handling depositions, client interviews, and witness preparation — where the record is evidence and the audio is privileged.
- Journalists working on legal stories — the source-protection logic from the journalist's playbook applies at higher stakes, because material may be subpoenaed.
- Investigators, compliance officers, ombudspersons — internal statements whose handling rules resemble legal ones even when no case exists yet.
- Court reporters' overflow — when a certified transcript isn't needed but a fast accurate working record is.
Why the upload path fails here, structurally
The ordinary transcription workflow — upload the file, get text back — assumes the recording is ordinary. Case material isn't: it contains admissions, allegations, strategy, and the identities of people who may not want their involvement known. Four properties of the cloud path create exposure that the local path eliminates:
- Third parties in the chain. A transcription vendor and its speech-recognition provider process the recording; whether that chain is discoverable, or breaks privilege expectations, is a question you've created and now have to answer. We map the same mechanics generally in what notetakers do with your audio.
- Retention you don't control. Vendor retention policies are written for uptime, not for case closure. "The matter settled; the recording is gone" is hard to promise when copies live in someone else's backups.
- Access surface. An account with the vendor is one more credential that reaches your case files — the same access-control point we make about where recordings are stored.
- Cross-border placement. Storage location is fixed by the vendor, not by your matter's jurisdiction.
The local path has one link: your machine. Recognition runs in your browser, the transcript lands in your browser profile, and deletion is a folder operation. That doesn't make the recording "secure" in some absolute sense — your laptop's disk encryption and your own discipline are the security — but it removes every party who isn't bound by your obligations.
Accuracy: where to verify, and why it's cheap here
Modern speech recognition on clear two-party audio is strong but not court-grade, and no tool should ship a legal quote unchecked. The verification burden is manageable because errors cluster predictably — names, numbers, terms of art, crosstalk — and because QuickSpeak keeps the audio next to the transcript: any quotable phrase is one click from its source recording. Before anything leaves your machine:
- Load the case vocabulary first. Party names, the judge, the statute numbers, defined terms, product names — the free custom-vocabulary list fixes their spelling in every future session, including live.
- Verify every quote and every number against the audio. Numbers are the classic silent error: "seventeen" vs "seventy" changes the case, and no spellcheck catches it.
- Label roles, not just names. Rename diarization labels to ATTORNEY / WITNESS / INTERPRETER — cleaner than names for a working record, and unambiguous in excerpting.
- Export the format the matter needs. .txt (free) for working notes; Markdown/JSON (Pro) for structured review workflows; the WAV stays paired for playback.
- Apply the retention rule. Local files mean the matter's own retention schedule governs — keep what the rules require, delete what they require deleting, and the deletion is real.
What this is not
Plainly: QuickSpeak is not a certified court-reporting service and produces no certified transcript. Where a matter requires a certified record, that's a licensed professional's job — a working transcription tool doesn't replace it. Recording rules also vary: consent requirements apply to depositions and statements as to any recording, and some proceedings restrict recording entirely — verify before the record button. This article is general information, not legal advice.
FAQ
Why can't deposition audio go to a cloud transcription service?
Because the recording contains privileged, case-sensitive material, and the upload inserts vendors into a chain your obligations may govern. On-device transcription keeps the whole process on your machine.
What accuracy do legal transcriptions require?
Exact wording for quotes, motions, and defined terms. Use machine output as a working draft and verify every quotable phrase against the audio — which stays paired locally.
Can the recording stay on my machine throughout?
Yes — capture, recognition, and storage all happen locally in QuickSpeak. No upload, no account; retention and deletion are yours.
Does it label who spoke?
On-device diarization separates speakers; you rename the labels to roles. Case vocabulary (parties, statutes, terms of art) goes into the free custom-vocabulary list for correct spelling in every session.
Read next
Sources & notes
- QuickSpeak features and limits (on-device WASM transcription and diarization, free tier scope, custom vocabulary, 500MB upload, Markdown/JSON export, pricing, 36 recognition languages): the shipping extension and usequickspeak.fyi, accessed November 13, 2026.
- Privilege, recording-consent, and retention statements here are general information, not legal advice; requirements vary by jurisdiction and matter. Verify with counsel.