Replace marked ranges of a recording with a bleep, with silence, with the same audio reversed, or with a muffled copy. Every range is placed by hand and can be set to the millisecond.
Mark the ranges
There are two ways to mark one. Drag a selection on the waveform, where the word is visible as a burst, and press Censor selection. Or drag straight across the lane under the waveform.
Each range appears as a band on both. On the lane, drag a band to move it and drag either edge to resize it. A band stops when it meets its neighbour, so two ranges never overlap.
Under the lane, each range has a row: Start and End in seconds, the replacement, Play, and Remove. Typed times are stored to 0.0001 s. The field arrows step by 0.01 s.
Play starts 0.75 s before the range and stops 0.75 s after it. A seam is judged from the audio on both sides of it, never from inside the range.
Finding the words is manual
Nothing here listens for speech. Locating words automatically takes a speech recognition model, and the page ships none.
The working method is to play the file, stop near the word, and drag a selection around the burst in the waveform. Loop that selection with the transport to check that it holds the whole word and nothing after it, then censor it. A short word runs a few tenths of a second, so expect ranges of roughly 0.2 to 0.6 s.
Four replacements
| Replacement | What is written | Recoverable |
|---|---|---|
| Bleep | A sine tone, 1000 Hz unless changed | No |
| Silence | Zeros | No |
| Reverse | The same samples in mirror order | Yes, by reversing again |
| Muffle | The same audio through a 24 dB per octave low-pass | Partly |
Bleep and Silence discard the original samples between the seam fades. Those are the two to use when the word must not come back out of the exported file.
Reverse and Muffle are editorial choices. A reversed word sounds like speech and keeps the sentence’s rhythm, which suits comedy edits. A muffled word sounds as if it came through a wall: with the cutoff at 300 Hz the consonants are gone and the vowels are a hum, yet a listener who knows the context can guess some words.
Bleep pitch and level
The tone is a plain sine. 1000 Hz is the broadcast convention; lower pitches read as softer and higher ones cut through small speakers.
With Match bleep to the voice on, the processor tracks how loud the recording runs and fixes the tone’s level at the moment each range begins. At 0 dB the tone carries the same RMS level as the voice before it; the default is -6 dB. The measurement rises in about 100 ms and falls over 3 s, so the breath before a word does not pull the bleep down.
With matching off, Bleep level is decibels below full scale. -12 dB is loud on most material.
Seams
A range that cut straight from speech to a tone would step the waveform, and a step is a click. Each edge is a raised-cosine crossfade between the original and its replacement, 10 ms by default and adjustable from 2 to 100 ms.
The fades sit inside the range. A range from 1.200 to 1.800 s changes nothing before 1.200 s, so the first few milliseconds inside it still carry a trace of the original on its way down. If the start of a word leaks, move the range start earlier by the fade length instead of shortening the fade.
Reversed ranges
A reversed range of up to 10 s is mirrored whole. A longer one is turned over in 10 s pieces, and the pieces fade to silence where they meet.
Reversal makes the live preview run late by the length of the longest reversed range, because the end of the range has to arrive before its beginning can be played. The export is aligned with the source.
Limits
- Up to 128 ranges, each at least 20 ms long.
- The export is re-encoded in the source’s format. A lossy source such as MP3 goes through a second lossy pass; start from WAV or FLAC when that matters.
- The censor is also a step in the audio workflow, where it can sit in front of a limiter. Ranges are drawn on this page; inside a chain they are carried as they are.