The bot finds obscene words in the speech, takes their timecodes and bleeps the right spots — with a tone or silence. No hunting for swearing on a timeline by hand, cutting the track and checking each fragment in an editing program.
Promo access unlocks every Flowmark tool for 30 days. Right inside Telegram and MAX.
An interview, a podcast, a recorded broadcast, a report, a short video for a channel — anywhere speech is already recorded but obscene language needs to come out before publishing. Usually that means opening an editing program, hunting for the words on a timeline and placing a tone or silence by hand.
Flowmark takes that tedious route off your plate: send the audio or video to the bot, choose censoring and get back a file with the detected words already covered.
Under the hood is transcription with word-level time alignment: the exact start and end of every word are known. So the bot covers precisely the right segment — it doesn't mute an extra second around it or cut the whole phrase. The same engine can simply turn speech into text — without censoring.
Before matching, the text is normalized — case, punctuation, Cyrillic and Latin are brought to a common form. On top of that, bigrams are checked: pairs of words that are each harmless alone but together form an obscene expression.
Before recognition the recording is always split into voice and background: music, street noise and beds are set aside, and a clean vocal track goes into the work. There is no “for speech” or “for music” mode to pick any more.
For Russian speech a second recognizer tuned for the language runs in parallel. It re-checks the recording and, when it finds a word the first algorithm misheard, moves it onto exact time boundaries. Fast, slurred speech and strong accents are now covered noticeably more cleanly.
A classic tone "bleep" over the words, or silence in the chosen intervals — your pick before you start.
Several thousand words with common typos and transliteration, plus a check on word pairs.
Works with both audio files and video — the sound track is processed, the picture stays as is.
In the Mini App you can exclude words that are fine for your team and add your own — slang, brands, anything. Plus a live phrase check with the production matcher.
The “Muted: N” button shows every trigger with context. A false positive? Exclude the word and rebuild the file for free — without repeated recognition.
Voice is always separated from music and background noise — no mode to pick. Only the vocals are censored, so the arrangement stays intact.
Besides the finished file, on request Flowmark assembles an FCP 7 XML — a standard interchange format. It's suitable for import into Premiere Pro and DaVinci Resolve; compatibility with other editors depends on the version and import settings. The XML already has the intervals laid out, with the chosen method — tone or silence — and exact timecodes matched to the source.
Alongside it, a separate track with that tone or silence is added to the archive, cut to the right timecodes. The editor opens the XML in their program, attaches the source — and gets the markup ready to go: re-voice, cover with a jingle, or leave it as is.
The same transcription engine turns speech into text, and right beside it in the bot — video compression and logo branding:
By word-level timecodes: the exact start and end of each word are known, so just the right segment is bleeped — not a second around it and not the whole phrase.
A tone over the words, or silence in the chosen intervals. One mode applies within a single task.
Russian and English — dictionaries of several thousand words with typos and transliteration, plus a check on word pairs.
Yes. On request an FCP 7 XML is assembled with the intervals and a replacement track. The format is suitable for import into Premiere Pro and DaVinci Resolve; compatibility with other programs depends on the version and import settings.
Automatic recognition isn't perfect. A missed word can be covered via the XML in an editing program, while a false positive can be excluded with the “Muted: N” button and the file rebuilt for free, without repeated recognition.
Yes. Each group has its own dictionary in the Mini App: exclusions from the base dictionary and your own added words. You can also check a phrase there and see exactly what would be muted.
Nothing: the recording is always split into vocals and instrumental first, censorship applies to the voice only, then the tracks are mixed back. There is no “For music” mode to choose any more — the separation is always on, and it also helps recognize speech under music and noise.
There's a free plan, and the promo link opens 30 days of extended access.
