Settings

The recording

MP3, WAV, M4A, OGG, FLAC or WebM, or a video, whose sound track is used. It is decoded in this tab and never leaves it.

Drop a recording here or choose it

How it works

A bleep or silence replaces the audio, so the words cannot be recovered from the result. From an MP4, WebM or MOV it takes the sound track and returns the sound only. The recording is processed in your browser and never uploaded.

The guide: Transcribe a recording without uploading it

Questions people ask about HUSH

What does each treatment do?

A bleep replaces the span with a steady 1 kHz tone, the sound television uses. Silence replaces it with nothing. A garble keeps a low, muffled version of the voice, filtered below 300 Hz and then ring-modulated, so the rhythm of speech is there but the words cannot be made out. Every edge has a ten millisecond fade just outside the span so the cut does not click, and every sample inside the span is treated in full.

Can the words be recovered from the result?

Not from a bleep or a silence: the original samples are not in the file, and the tone or the silence is all that is written there. A garble is filtered so hard that the consonants and the vowel shapes are gone before the ring modulator scrambles what is left, which is far beyond what a de-noising tool undoes; still, if the recording must never be recoverable, silence is the safe choice. Nothing here hides the original behind the treatment: the WAV is written from the treated samples only.

Why is the output a WAV?

Because no MP3 or AAC encoder is shipped on this site, and a WAV of 16-bit PCM is what every player, editor and phone opens. It is written at the sample rate and channel count the recording was decoded at, so nothing is resampled or folded to mono. It is larger than an MP3 of the same length: about ten megabytes a minute in stereo at 44.1 kHz. If you need an MP3, convert the WAV afterwards in any converter.

Does it work on a video?

It takes the sound track out of an MP4, WebM or MOV that your browser can decode, and the result is a WAV of the sound only. The picture is not touched and is not in the result. To bleep the sound in the video itself, use this WAV as the replacement track in a video editor.

How do I mark a word precisely?

Zoom in with the button, or hold Ctrl and turn the wheel over the waveform, until the word is wide enough to drag across. Once a span is selected its edges can be dragged, and the From and To boxes take an exact time. If you have a transcript with timestamps, such as the SRT that SCRIBE writes, paste it into the list and every line with two times becomes a span; delete the ones you do not want.

What can I paste into the list?

One span a line, as a start and an end time, like 0:01.250 0:01.900 or 1.25 1.9. Subtitle lines with an arrow, like 00:00:01,250 --> 00:00:01,900, work as they are, so an SRT file can be pasted whole; the index lines and the text lines are left out. JSON works too: a list of two-number pairs, or of objects with start and end, or the chunks with timestamps that a Whisper-style transcriber writes. A word on the line such as bleep, silence or garble sets the treatment for that span.

Is anything uploaded?

No. The recording is decoded by your browser, the result is rendered by your browser, and the WAV is written by your browser. The page works with the network switched off, and nothing is kept once the tab is closed.

What does it not do?

It does not transcribe: SCRIBE does that, and its timestamps can be pasted here. It does not find the words for you, change the pitch of a voice, or write MP3. Very long recordings need memory: an hour in stereo is about a gigabyte of samples in the browser while it is being worked on, so on a phone keep to shorter files.

Doing this often? OBSCURA Sound edits the whole recording: long silences shortened, hiss and hum reduced, the level evened out, and the result set to a podcast's loudness. Everything is US$5 a month or US$48 a year.