Drop in a recording and it gets transcribed, phone numbers and card numbers and emails and ID numbers read aloud get flagged as they're spoken, and you click any word to redact a name or anything else by hand. Transcription, review, redaction, and export all run on your device, and the recording itself is never uploaded anywhere.
As the transcript streams in, phone numbers, card and bank numbers, email addresses, and ID numbers spoken aloud get flagged on their own, including numbers read out digit by digit ("four two four two…"), which get normalized to digits before matching.
Names are usually the whole reason someone wants a recording redacted, and reliably recognizing a spoken name is a much harder problem than recognizing a phone number, so it doesn't pretend to. Click a word, or drag across a phrase, and choose bleep or mute. Click a single word and "redact all occurrences" flags every instance at once: a multi-word selection doesn't offer that, since matching a whole phrase everywhere it's spoken reliably is a different problem. A custom term list flags a name or phrase automatically from then on, without you re-clicking it each time it comes up.
Set the mode per redaction, or switch a whole group at once.
Two checks keep that from being a leap of faith. Every redaction gets its own preview button, so you hear exactly what's being removed before you commit to it. Try to export with redactions you haven't listened to yet and you'll be asked to confirm first. After you export, the finished file gets decoded and checked span by span: a muted span has to measure out near-silent, a bleeped span has to measure out as dominated by the tone, and if either check fails on a specific span, the export is held back and that span is called out rather than handed to you with a hidden leak in it. (A recording too long to decode and check this way, past 45 minutes on a computer, 10 on a phone, skips that automated pass with a plain notice, and gets a spot-check playback of the exported file instead.)
Nothing here is saved between visits: close the tab and the recording, transcript, and flags are gone. An optional export log (times, redaction type, and a count, without the matched words unless you separately ask for them) is available if you need to attest to what was removed.
Audio, or a video file with an audio track: you'll get back just the redacted audio. Video output isn't in yet.
Speech recognition turns the recording into a transcript as it goes, with automatic flags appearing the moment they're found, so a long file is reviewable before it's even finished transcribing.
WAV is lossless (the safer pick for court or compliance use), MP3 is smaller and more universal. Either way, the file's own metadata (recording-app tags, GPS, embedded titles) is stripped from the export by default.
Redact a patient's name and details from a therapy or telehealth recording before it goes to a transcription service, a supervisor, or a training set: no vendor contract or Business Associate Agreement required.
Prepare an audio exhibit without exposing a privileged conversation or a third party's details: redact a client call or deposition recording before it enters the record.
Scrub a card number or account detail read aloud on a support call before the recording is stored or reviewed: the PCI-DSS pressure, handled without a vendor contract.
Anonymize interview audio for IRB compliance before archiving or sharing a dataset: a participant's name is gone before the file ever leaves your laptop.
Protect a source's identity before publication: bleep a name or a location without re-recording the segment.