TRANSCRIPTION FOR JOURNALISTS

Transcription for Journalists — Keep Source Audio Off the Cloud

Every cloud transcription service is a third legal custodian of your reporting. Under the Stored Communications Act the government can compel them for your tape — often without notifying you. VocalFuse transcribes with local Whisper on your Windows PC: no upload, no bot, no vendor account — $5/mo or $79 one-time.

Transcribe locally with VocalFuse The local workflow The subpoena problem

The subpoena problem with cloud transcription

You promised a source confidentiality. A shield law defends that promise in court — but the moment your interview audio sits in a vendor's S3 bucket, the fight is no longer only yours. The government can serve the provider directly, and the provider faces real penalties for refusing.

The third-party doctrine. Material you voluntarily hand to a service gets weaker Fourth Amendment protection — the doctrine behind Smith v. Maryland (1979) is why cloud-held content is treated differently from the laptop in your bag.
18 U.S.C. § 2703. Content stored ≤180 days generally requires a warrant; older content historically could be reached with a grand-jury or administrative subpoena or a § 2703(d) court order on "specific and articulable facts" — no probable cause. Long retention is long exposure.
Delayed notice. Under § 2705(b) a court can gag the provider so you learn about the disclosure up to 90 days later — renewable. A source can be identified before you ever know a subpoena landed.
Training and retention. Several services reserve rights to use uploaded audio to improve models, and retain recordings until you delete them. De-identification is not deletion.

None of this means a cloud tool will be abused — it means the architecture puts the decisive copy of your source's voice where you have no control and sometimes no notice.

Cloud vs local: where your source tape actually lives

Question	Cloud notetaker (Otter, Rev, Descript, Trint)	Local transcription (VocalFuse)
Who holds the audio?	The vendor's cloud storage — they are the custodian a subpoena reaches	Your PC only. No provider copy exists to compel
Can you be notified of every demand?	Not always — § 2705(b) gag orders can delay notice 90 days, renewable	No third party to serve. Any demand targets you, where shield-law process applies
AI training on your tape?	Some services reserve rights to train on uploaded content (litigation active in 2026)	None — transcription runs on-device, nothing is sent anywhere
Voiceprints of your source?	Speaker profiles are biometric models of a voice — deletion is a separate fight	No voiceprints, no speaker biometrics, no cloud speech API
Retention you control?	Vendor's policy plus your settings; deleted ≠ gone from backups	Yours entirely — delete the file, it is gone

What shield laws do — and what they can't

What they do

40 states + D.C. have shield statutes; most federal circuits recognize a qualified reporter's privilege
16 states + D.C. protect confidential sources absolutely — a subpoena to you fails there
DOJ regulations (strengthened 2021) restrict federal prosecutors' news-media subpoenas

What they don't

No federal shield law — the PRESS Act has passed the House but never cleared the Senate
The privilege vests in you — it cannot stop a § 2703 process served on your provider
The Privacy Protection Act guards newsroom materials from searches — not copies held by a vendor
Agency workarounds (e.g., 19 U.S.C. § 1509 customs summons) route around judicial process entirely

The Reporters Committee for Freedom of the Press maintains the definitive state-by-state privilege compendium. Know your state's scope before you need it.

Recent incidents worth knowing

DOJ vs. Times reporters (2026)

Secret subpoenas to telecoms sought phone records of multiple New York Times reporters in July 2026; DOJ withdrew them on July 23 after the Times moved to quash. Documented on the U.S. Press Freedom Tracker.

DHS customs summons (2026)

DHS used 19 U.S.C. § 1509 — a customs provision with no judicial sign-off — to obtain six months of a Minneapolis journalist's call and text logs from T-Mobile, with instructions to keep the demand secret.

Brewer v. Otter.ai (2025–)

A consolidated federal class action alleges Otter's meeting bot recorded participants without consent and trained on the audio; a 2024 OtterPilot incident leaked patient data in an Ontario hospital.

Guidance from Freedom of the Press Foundation and GIJN reaches the same conclusion: minimize what sensitive tape ever reaches a third party.

A local workflow for protected-source interviews

1. Record on your device

Use a recorder you control — your phone's voice memo app, a field recorder, or your PC's microphone. Confirm your state's consent rule first; get consent on the record when in doubt.

2. Transcribe locally

Run the audio through a local Whisper model (VocalFuse on Windows). No upload step exists, so no speech API, no vendor server, and no account is involved in the transcription itself.

3. Pseudonymize what leaves your machine

Before a transcript goes to an editor, a co-author, or a draft: replace source names and identifying details with codes. Keep the key separately, encrypted, or not at all.

4. Keep retention minimal

Verify the transcript, export what the story needs, then decide deliberately whether the raw tape still earns its risk. Delete what you would not hand over willingly — a file that does not exist cannot be produced.

This page is operational guidance, not legal advice — for a live subpoena or a high-risk source, involve your newsroom's counsel and the RCFP's legal defense resources.

VocalFuse: transcription that physically cannot phone home

Local Whisper on Windows

Speech recognition runs on your GPU/CPU with a local Whisper model. Drop in an interview file or record live — the audio never leaves the machine, and there is no bot to join anyone's call.

No account required to transcribe

Core transcription needs no signup. There is no server-side copy of your audio, so there is nothing for a § 2703 demand to reach — the tape exists only where you put it.

Pro notes are opt-in

Pro ($10/mo) can draft structured notes and summaries from the transcript. Sync is optional and covers text only — for protected-source work, keep sync off and everything stays local.

Flat pricing

$5/mo or $79 one-time — no per-minute metering that penalizes long interviews, no AI credits that run out mid-investigation. See VocalFuse features.

Get VocalFuse for Windows

Transcription for journalists — FAQ

What is the most secure transcription for journalists?

Transcription that never leaves your machine. Cloud services (Otter, Rev, Descript, Trint) hold your source audio on their servers, which makes the provider — not you — the legal custodian of your tape under the Stored Communications Act. A local Whisper setup transcribes on your own Windows PC: there is no upload, no vendor account, and nothing on anyone else's server to hand over.

Can police subpoena my recordings from Otter.ai or another cloud service?

Yes. Under 18 U.S.C. § 2703 a provider can be compelled to produce stored content — a warrant for content under 180 days old, and historically a subpoena or § 2703(d) court order ("specific and articulable facts") for older material. Providers can be gagged under § 2705(b) so you are not notified for 90 days, renewable. Your shield-law argument happens in court later; the provider's compliance can happen without you ever knowing.

Do reporter shield laws cover cloud-stored recordings?

Not the copy the provider holds. Forty states plus D.C. have shield statutes and most federal circuits recognize a qualified privilege — but the privilege vests in the journalist, and it protects you from being compelled in court. It does not stop the government from serving process on the cloud provider that stores your audio. The Privacy Protection Act protects newsroom materials from searches; it does not cover a vendor's copy. The only copy a subpoena cannot reach is one that does not exist outside your machine.

Is it legal to record an interview with a source?

It depends on your state's consent rule: one-party-consent states let you record a conversation you are part of; all-party-consent states (like California) require everyone to agree. Federal law requires one-party consent. When reporting crosses state lines or the source is in an all-party state, get consent on the record — it also builds trust. This page is not legal advice; confirm your jurisdiction's rules before recording.

How do I transcribe an interview without uploading the audio?

Record locally, then run the file through a local Whisper model on your own machine — VocalFuse does this on Windows with no upload step and no account required for transcription. Verify the transcript on screen, export what you need, and keep the raw tape under your own retention rules. Nothing is sent to a speech API, so no third party ever holds your source's voice.

Does VocalFuse upload audio or create voiceprints?

No. VocalFuse transcribes with a local Whisper model on your Windows PC — no meeting bot, no cloud speech API, no speaker voiceprints. On the optional Pro plan, text transcripts and summaries can sync to your account; for protected-source work keep sync off and everything stays on the machine. Flat pricing: $5/mo or $79 one-time.

What happened with Otter.ai and journalists?

A federal class action (Brewer v. Otter.ai, filed August 2025) alleges Otter recorded conversations without every participant's consent and used the audio for AI training; a separate 2024 incident saw an OtterPilot bot transcribe patient data in an Ontario hospital meeting. Freedom of the Press Foundation and GIJN both advise minimizing cloud transcription for sensitive tape. The risk is not malice — it is that automated cloud workflows handle sensitive material in ways you did not intend.

Has the government actually sought journalists' records from providers?

Yes. The U.S. Press Freedom Tracker documents DOJ subpoenas in July 2026 for phone records of multiple New York Times reporters (withdrawn July 23, 2026), and DHS used a 19 U.S.C. § 1509 administrative summons to obtain six months of a Minnesota journalist's call and text logs — after a judge had already pushed back on similar requests. The pattern is consistent: authorities go to the provider first, and secret process is part of the design.