Say goodbye to bad words in your videos

Trying it is completely free, and the try is the real thing: drop in your own file, and the on-device AI transcribes it, highlights every word from your list, and lets you hear each bleep in place — no account, no watermark, no upload. Keeping the censored file is the paid step: $4.99 buys a 24-hour Day Pass with unlimited exports and every Pro feature, no subscription. Regulars pay $3.99/month or $24/year, and $79 buys it outright forever. We flipped from a traditional free tier to free-preview-plus-Day-Pass so you never pay to find out whether it works on your audio — you pay after you have watched it work. Cancel anytime, and there is a 14-day no-questions refund on every purchase.

No. CensorMate is a browser-based profanity censor whose AI runs on-device: the speech-recognition model (OpenAI’s Whisper) downloads to your browser once, and after that transcription and censoring happen entirely on your own hardware. Your audio and video files never leave your computer or phone — not to our servers, not to any cloud AI, not to anyone. There is no account, and we could not see your files even if we wanted to, because no upload path exists in the product. This matters beyond privacy preference: recordings containing other people’s voices, children, students, or client material often cannot be handed to a cloud service at all, and on-device processing is what makes censoring them defensible.

Yes to both, and neither requires talking anyone out of it. Cancel anytime yourself via the "manage subscription" link in your receipt email from Polar (our payment provider) — Pro keeps working until the end of the period you already paid for, then simply drops back to the free preview, keeping your keywords and settings. For refunds, email support@censormate.com within 14 days of purchase and you get a full, no-questions refund processed back to your original payment method, usually within 5–10 business days depending on your bank. Subscriptions are handled by Polar as the merchant of record, so receipts, taxes, and card handling are all managed there rather than by us.

Yes — open censormate.com/app in your phone’s browser (iPhone Safari or Android Chrome) and it just works: no app install, no account. Phones run a lighter on-device speech-recognition engine tuned for mobile hardware, so the AI still runs entirely on your device and nothing is uploaded — the same privacy promise as desktop. The realistic expectations: short clips (roughly a few minutes) work great, which covers the TikTok/Reels/Shorts use case the phone flow is built for; very long or high-resolution videos are happier on a computer, simply because phone browsers get a fraction of the memory. The one-time AI model download is about 75MB on phones, cached after the first run. Everything else — transcript review, bleep fine-tuning on the timeline, captions — works the same.

Audio: MP3, WAV, M4A, OGG and most common variants. Video: MP4, WebM, MOV, MKV — including iPhone .MOV recordings straight off the camera roll. For video exports, the picture is genuinely untouched: CensorMate copies the video stream bit-for-bit and rebuilds only the audio track with your bleeps, so there is no re-encoding and no generation loss — the exported picture is identical to the original, not "visually similar." If your browser can’t decode a file’s audio directly, a built-in FFmpeg fallback extracts it, which is how the trickier containers like MKV work. Alongside the censored file you can also download matching starred captions (.srt or .vtt) and a plain-text transcript.

CensorMate uses OpenAI’s Whisper speech-recognition model with word-level timestamps, plus boundary refinement so each bleep lands exactly on its word rather than clipping the one before or after it. Accuracy in practice depends on audio quality and the model you pick: the Balanced model handles typical creator audio well, and Pro’s most-accurate model helps with fast talkers, accents, and noisy rooms. Crucially, you are never asked to trust the AI blindly: the full transcript appears with every match highlighted before anything is censored, you can toggle any word on or off with a click, and each bleep can be nudged on a waveform timeline to the centisecond. The AI does the finding; you keep the final say.

21 languages, including English, Spanish, French, German, Portuguese, Italian, Dutch, Polish, Russian, Ukrainian, Japanese, Korean, and Chinese. The same on-device Whisper model family powers all of them, so nothing changes in the workflow: pick your language, and transcription, keyword matching, and censoring work the same way. Two honest caveats: recognition accuracy is strongest in English and the other high-resource languages, and keyword matching is exact-text based — so for languages with heavy inflection you may want to list a few word forms (the partial-match option helps by catching your term inside longer words). The transcript review step matters a little more outside English; it is also what catches anything the model heard slightly differently.

Any word or phrase — profanity is just the default use case. Creators censor brand names before sponsorships clear, spoilers in reaction content, and their kids’ names in family vlogs; HR and legal teams censor third-party names and personal details; teachers make classroom-safe versions of real-world media. Multi-word phrases work ("123 Maple Street"), matching ignores case and punctuation, and you can upload an entire keyword list as a .txt file or save lists as reusable presets. CensorMate also flags patterns that look like phone numbers, email addresses, and SSN- or card-shaped numbers spoken aloud, and offers to censor them in one click — useful for recordings where the risky content is personal information rather than language.