Guide

Transcribing confidential interviews without exposing your source or client

Updated July 15, 2026·8 min read

Some conversations carry an obligation heavier than convenience: a journalist protecting a source, a lawyer under privilege, an HR professional in a sensitive case, a researcher who promised anonymity. Transcription makes that work manageable, but most tools quietly break the promise by sending the audio to someone else's server. We built Fieldnoter for this kind of material, so read this as coming from the people who make one of the on-device options. It covers what that upload actually means, what to settle before you press record, and a full workflow that keeps the material on hardware you control.

What uploading means for these conversations

When interview audio is processed in the cloud, a third party technically holds a copy of the most sensitive material you have. For most professions that raises specific problems:

• Journalists: source protection weakens the moment a subpoena-able company holds the recording. This is not hypothetical; it is why newsroom security guides warn against cloud transcription for sensitive material.
• Lawyers: privileged conversations that pass through a third-party processor create questions you do not want to answer later, and many firms prohibit it outright.
• HR and works councils: a grievance conversation on a US server is a data protection incident waiting to be described that way.
• Researchers: ethics boards increasingly ask where interview audio is processed, and "a cloud service" prolongs approval or sinks it.

To be fair to the cloud tools: they offer things a phone cannot, like team libraries, meeting bots that join calls, and larger models for the noisiest audio. For everyday meeting notes those are reasonable trades. For material where a single copy in the wrong place is a professional failure, they are the wrong default.

Before you press record

None of this is legal advice. Recording law differs by country, and your professional obligations may be stricter than the law. Whatever tool you use, four things are worth settling before any sensitive interview:

• Consent. Know whether your jurisdiction requires all parties to agree to a recording, and what your profession expects on top of that. For most interview work, asking on the record is standard practice anyway.
• Retention. Agree, with yourself or your organization, how long the audio and transcript will exist and when they get deleted. A recording you no longer hold cannot leak.
• Anonymity promises. If you promised a source or participant anonymity, decide before recording how names will be handled in the transcript and in anything you export.
• Your organization's rules. Newsrooms, law firms, HR departments, and ethics boards often have recording policies of their own. Check them first; a tool choice does not override them.

The alternative: keep processing on your device

Modern iPhones can run speech recognition, speaker separation, and summarization locally. Practically, that changes the terms:

• The audio exists in exactly one place, on hardware you control, protected by the device's encryption.
• There is no company that can be compelled, breached, or acquired with your recordings on its servers.
• You can verify the claim instead of trusting it: transcribe with the phone in airplane mode.

How this works technically: on-device transcription explained. One caveat belongs here, and we would rather state it than let you find it: an on-device app still makes a single network call, a one-time download of the speech model at first setup. After that it runs offline, and you can check it yourself:

The workflow, step by step

The full loop in Fieldnoter, from conversation to archive:

1. Record or import. Record live with the phone in the room, or import audio or video you already have. Both go through the same local pipeline.
2. Pick the interview template. Fieldnoter includes templates per type of conversation, including interview, legal, and 1-on-1. The template shapes what the summary pays attention to.
3. Label the speakers. Speaker separation splits the transcript by voice. Rename Speaker 1 to a real name with a tap, or deliberately leave the labels generic for material where even the transcript is sensitive.
4. Summarise and pull out follow-ups. The on-device summary gives you the shape of the conversation; action items can go straight to Reminders.
5. File it under a client or project. Recordings are organized by client or project inside the app, which keeps a long-running investigation or case together.
6. Export to the format your archive needs. TXT for reading, SRT or VTT for anything with video, JSON or CSV for structured archives. The table below has the details.

Importing an existing recording

Not every interview happens with your phone on the table. Perhaps you use a dedicated recorder, received audio from a colleague, or have the recording of a video call. Fieldnoter treats import as an equal citizen to live recording: share the file into the app or pick it from Files, and it runs through the same local pipeline. Video files work too; the app uses the audio track.

The point for confidential work: a recording that already exists does not need to be uploaded anywhere to become a transcript. Processing happens on the phone regardless of where the audio came from. The one thing import cannot fix is the past of the file itself; if the recording was made with a cloud service, a copy may already exist on that service's servers.

Export formats and what they are for

An archive you cannot read in ten years is not an archive. Fieldnoter exports to five plain, documented formats:

One caution: the export is where your control ends. A transcript pasted into a cloud document reintroduces exactly the third party you avoided during processing. Match the destination to the sensitivity of the material.

Practical notes for sensitive work

• Consent and law first. On-device processing protects the recording; it does not replace the rules about making one. Know your jurisdiction and your profession's requirements before pressing record.
• Lock the app separately. A phone gets handed around more often than you think. Fieldnoter supports a Face ID lock on top of the device passcode.
• Mind your exports. The private part is only as private as where you send the transcript afterwards. Exporting into a cloud document reintroduces the third party you avoided.
• Names can stay out of it. Speaker labels do not need real names for material where even the transcript is sensitive; leaving it at "Speaker 1" is a legitimate choice.
• Delete on schedule. When your retention period ends, delete the recording in the app. Because there is no server copy, deleting it there is deleting it everywhere.

Frequently asked questions