Your audio never leaves your device
Transcribe confidential interviews with the audio kept on your device
Turn recorded interviews into text without sending them to a cloud service. Everything runs in your browser, so a source recording or a research interview stays private and under your control.
Takes a little extra time, and your audio still never leaves your device. Labels go by how voices sound, so treat them as a best effort: one person can be split across two or three labels, similar voices can be merged, and someone joining partway through may be folded into a speaker already talking. Giving a number usually helps. Check them before relying on them.
Specifying this number improves accuracy. Count people who really take part, whenever they join. Someone who only says a word or two is better left out.
Timestamps each word instead of each phrase, so short interjections like "Yeah" go to the right person. On a busy four-person recording this corrected about 1 word in 5; it changes little when people talk in long turns. Adds roughly 40% to the time on a computer, and almost nothing on a phone.
Phone audio loses the high frequencies that separate consonants, so they are put back before transcribing. Only phone-quality recordings are treated, and you are told when it happens. Also even out volume is for one case: someone far from the microphone and much quieter throughout. It cost accuracy on other recordings, so use it only if a quiet person came out badly.
Worth filling in. The speech model knows ordinary English, not your subject, so a word it has never heard becomes whatever sounds closest, and it usually gets that word wrong every single time it comes up. A sentence is enough, and the names and jargon in it are picked out and watched for. It cannot guess a word you do not mention, so name the people and the terms that matter.
Free, no signup. The first time you use it, it takes a minute or two to set up. After that setup is quick. Works with MP3, WAV, M4A, OGG, FLAC and AMR, and long recordings are handled in parts.
Why this matters for interviews
Interview audio is uniquely exposing. Unlike a transcript, a recording carries the source's actual voice, which is identifying on its own. It carries the off-the-record aside before the interview properly started, the moment they named someone else, and the background sounds that place them somewhere specific. A transcript can be redacted. A recording cannot be un-heard.
Uploading that file to a transcription service creates a copy in a system with its own legal exposure, its own breach history, and its own obligations to respond to lawful requests, none of which you control or would necessarily be told about. MeetingScribe transcribes without uploading, so the only copy of the audio is the one you already have.
Where this fits in a source-protection chain
It is worth being precise, because overstating this would be its own kind of harm. Removing the cloud copy closes a real gap and leaves others open:
- Closed: a vendor holding your audio, a vendor breach exposing it, a vendor receiving a request for it, and audio being retained or used for model training under terms you skimmed.
- Still open: the recorder itself, the laptop, backups that sync automatically, the transcript once you share it, and anything you paste into another service later.
The practical implication is that on-device transcription is worth pairing with full-disk encryption, an intentional decision about whether your working folder syncs to a cloud drive, and a habit of deleting audio once the transcript is verified.
A workflow for interviews you intend to publish from
- Transfer from the recorder, and stop there. Copy the file to your working machine and transcribe it on that machine. Do not stage it on a shared drive first.
- Turn timestamps on. This is the single most useful setting for interview work. Verification later depends on being able to find the moment in seconds.
- Split anything over about an hour into parts named in order. Transcription itself is chunked automatically, but a very long file still has to be decoded into memory in one go, and joined parts are seamless.
- Rename speakers immediately while you still remember who was who. Two months later, SPEAKER_01 is a research problem.
- Verify before you quote.Treat every quotable line as unconfirmed until you have listened to it. Speech models are confidently wrong about names, numbers, and negations, and a dropped "not" changes a story.
- Keep a redacted working copy. Replace identifying names with role labels in the version you carry around, edit, and share with an editor, and keep the unredacted one in one place only.
For qualitative researchers
The same properties that suit journalism suit qualitative research, often for ethics-approval reasons rather than legal ones. Many ethics protocols commit to processing data only on approved systems, and a browser-based transcriber that never transmits audio is straightforward to describe in a data management plan: there is no third-party processor to name, because there is no processing off your device.
- Anonymize during transcription, not after. The in-page editor is where pseudonymization is cheapest. Doing it before the transcript is saved means the identifying version never exists as a file.
- Plain text imports everywhere. The .txt export loads into the common qualitative analysis packages without conversion, so coding can start immediately.
- Speaker-labeled export preserves turn structure, which matters if your analysis pays attention to who introduced a topic rather than only what was said.
- Record your method. For transparency, note the model size used and that transcripts were machine-generated and human-corrected. That is an honest description of what the data is.
Getting usable audio in the field
Field conditions are harder than any meeting room, and the transcript can only be as good as the recording:
- Cafes are the worst common choice. Espresso machines and background music sit in the same frequency range as speech and confuse both transcription and speaker separation.
- Put the recorder closer to your source than to yourself. Their words are the ones you need verbatim.
- Record WAV if your device offers it. Heavily compressed formats lose detail that the model needs, and storage is cheap.
- Choose the small model for field audio. This is the situation the extra accuracy exists for.
- Say the date and the participant reference at the top of the recording. It survives file renaming and lands at the top of the transcript.
Consent and rights
You are responsible for having the right to record and transcribe an interview, and recording laws vary considerably by location and by where each participant is. Our guide to recording law is general background rather than advice for a specific situation. MeetingScribe is a private way to convert audio you are permitted to use into text, not a judgment about whether a given recording was lawfully made.
Last reviewed: August 2026
Frequently asked questions
- Can I protect a source this way?
- It closes one specific gap: there is no cloud copy to be breached, subpoenaed, or handed over by a vendor. It does not protect a compromised laptop, an unencrypted drive, or a transcript you later email. Source protection is a chain, and this is one link in it.
- Does it handle two people talking?
- Yes. It separates speakers and lets you rename them, so a transcript clearly shows interviewer and interviewee. Accuracy is best with one to three voices and degrades when people talk over each other.
- Can I verify a quote before publishing it?
- Turn timestamps on before transcribing. Every passage then carries the time it came from, so you can jump straight to that point in the audio and confirm the wording rather than trusting the transcript.
- Does it work for field recordings?
- Yes, for MP3, WAV, M4A, OGG, FLAC, and AMR, which covers most handheld recorders and phone voice memos, including AMR files from older phone call recorders. The AMR decoder is loaded into your browser and runs locally, so those files stay on your device too.