Bleep Out Words in an Audio File

How to Bleep Out Words in an Audio File

  1. 1

    Download the listening model

    Finding the words means listening to the recording, which takes a speech recognition model — about 290 MB, fetched once and then cached for offline use.

  2. 2

    Type your word list

    Up to 50 entries, separated by commas or spaces. An entry with spaces in it, written between commas, is matched as a phrase. Signed in, you can keep one list against your account so it is there next time.

  3. 3

    Choose what goes over them

    The standard beep, silence, or a sound file of your own up to ten seconds long. Under Advanced settings you can change the pitch and level of the beep, and how much air it leaves around each word.

  4. 4

    Add your files and bleep

    Drop in one recording or a batch. Each result says how many words were covered and at what time, so you can check the pass without listening to all of it, and downloads as MP3 or WAV.

What this is

Censoring a recording by hand means listening to all of it, writing down the timecodes, and drawing a tone over each one. This does that pass for you: type the words you do not want in the programme, and every place they are said is found and covered — with the 1 kHz tone broadcasters use, with silence, or with a short sound of your own that plays for exactly as long as the word did. It finds the words by transcribing the recording with OpenAI's Whisper, running on your own machine, so nothing is uploaded and no service ever sees what was said. There is no built-in list of profanity, deliberately: it bleeps the words you list and no others, because whose list it would otherwise be is not a question a tool should be answering for you. Useful for podcasts going out on a clean feed, radio edits, classroom and training material, family versions of a stand-up set, and clips headed for a platform with rules about language.

How it works

Two different jobs, one button. Finding the words is Whisper, OpenAI's speech recognition model, run in your browser through Transformers.js and ONNX Runtime — the same model the transcription tool here uses, asked for word-level timestamps rather than sentence-level ones. Those come from the decoder's cross-attention aligned against the audio, which puts each word within a few tens of milliseconds of where it actually is; the "air around each word" setting exists to cover exactly that margin, so a bleep never lets the first consonant through. Covering the word is not AI at all: the samples inside the range are replaced with a sine wave at the pitch you choose, with silence, or with your own sound looped or trimmed to fit, and the audio either side is ramped down into it over four milliseconds so the join does not click. Matching is exact and whole-word — "damn" does not catch "damned" — because a bleep over a word nobody said is a hole in a sentence that used to make sense.

Credits

The research this tool runs on, credited in full — with BibTeX and licence notices — on the credits page.

  • Whisper

    MIT
    Alec Radford, Jong Wook Kim, Tao Xu, Greg Brockman, Christine McLeavey, and Ilya Sutskever. 2022. Robust Speech Recognition via Large-Scale Weak Supervision. arXiv:2212.04356.Read the paper

  • Transformers.js

    Apache-2.0
    Joshua Lochner and the Hugging Face team. Transformers.js: State-of-the-art machine learning for the web.

  • ONNX Runtime

    MIT
    ONNX Runtime developers. 2021. ONNX Runtime: Cross-platform accelerated machine learning.

Supported formats

WAV, MP3, M4A, AAC, OGG and FLAC. Files are decoded by your browser, so anything it can play will work here. Output is the same recording with the listed words covered, as an MP3 at the bitrate you choose or as an uncompressed WAV.

Frequently asked questions

Is my audio uploaded anywhere?
+Is my audio uploaded anywhere?
No. Everything runs inside your browser using the Web Audio API, so the file never leaves your device. You can prove it: load the page, turn off your Wi-Fi, and the tool still works.
Does it come with a list of swear words?
+Does it come with a list of swear words?
No, and that is deliberate. It bleeps the words you type and no others. A built-in list would be somebody's judgement about what counts as profanity — which varies by country, audience and broadcaster — quietly applied to your recording. Typing your own takes a minute and is the only list that is actually right for what you are making.
How accurate is it?
+How accurate is it?
It is as accurate as the transcription underneath it, which is very good on clear speech and less good with heavy accents, crosstalk or noise. Two things help: pick the larger Whisper model under Advanced settings, and clean the recording up with the noise removal tool first. Every run lists what it covered and when, so a missed word is something you can see rather than something you have to catch on playback.
Can I use my own sound instead of a beep?
+Can I use my own sound instead of a beep?
Yes — any audio file up to ten seconds. It is fitted to each word: longer than the word it is cut off, shorter and it repeats until the word is over. The file stays in your browser tab and is never uploaded or stored, which does mean a saved chain that used one will ask for it again next time.
Will it catch plurals and variations?
+Will it catch plurals and variations?
Only the ones you list. Matching is on whole words, so "damn" does not catch "damned" and — more importantly — "ass" does not catch "class". Add the variants you want caught; fifty entries is plenty of room for them.
Can I bleep as part of a bigger edit?
+Can I bleep as part of a bigger edit?
Yes — Bleep Words is also a block in Chain Tools, so a single pass can strip the background noise, censor the words, level the result and write out an MP3. Put it early in the chain, before anything that changes the timing.
Does this use AI?
+Does this use AI?
Half of it. Finding the words is OpenAI's Whisper, a speech recognition model, running on your own device — that part is genuinely AI and is why the tool needs a download before it can start. Covering the words is plain signal processing: a sine wave, silence, or your own file written over the samples. Neither half sends your recording anywhere.
Do I need an account?
+Do I need an account?
No. The free tools work straight away with no sign-up, up to 15 minutes of audio per tool per day. An account is only needed if you subscribe to remove the limit.
Is it free?
+Is it free?
Yes. Every tool on this page is free to use, with a daily limit on total audio processed. There is no watermark and no trial period.

Other free audio tools