SHOOSH removes explicit words from any song in 20 to 35 seconds, keeping the music intact using AI stem separation and word-level transcription.
Drop a song. AI does the dirty work. You pick the cleanup style. Out comes a polished, broadcast-ready track in under a minute.
AI splits your track into vocals + instrumental in seconds. The music keeps playing clean under every shoosh.
How it works: a Demucs neural network isolates the vocal stem from the rest of the mix. We only edit the vocal so the instrumental stays full-quality, untouched.
Bonus: opt in to extract all 6 stems (bass, drums, vocals, guitar, piano, other) as a background job. Costs +1 credit; doesn't slow your shoosh down.
200×-realtime speech-to-text gives you word-level timestamps. Every lyric pinpointed to the millisecond.
How it works: Groq Whisper Turbo runs the transcription in 1–3 seconds for a four-minute song. No waiting around for batch jobs.
Every word arrives with its start / end in seconds, accurate enough to surgically replace a single syllable.
Every word rated for offensiveness. Pick a censor method. Threshold slider tunes everything else.
How it works: a fast LLM rates each word's intensity 0 to 1. The threshold slider, or the MPAA-style rating cards (G / PG / PG-13 / R / NC-17), controls how aggressive the cleanup is.
Click any flagged word to skip it. Click any normal word to add it manually. You're always in control.
Each method runs in seconds. Hover any card to see what it sounds like.
Classic
The universal TV-censor tone. Instantly recognizable. Safe for any context, never surprises a listener.
Silent
Dead silence. The instrumental plays right through. The word is just gone. Cleanest possible result.
Hip-hop
A quick vinyl scratch over the word. Hip-hop heritage. Adds character without disrupting flow.
Subtle
The vocal plays backwards just for that word. Weird in the best way. Often goes unnoticed.
Demon voice
Drops the vocal two octaves for the duration of the word. Unintelligible. Slightly menacing. Iconic.
Wind-down
A turntable wind-down effect on the vocal. Like the record stopped for a beat. Cinematic.
Your sound
Drop your own MP3: airhorn, meme clip, custom sting, anything. We splice it in at every flagged word.
Anywhere clean audio matters. Click the dots to flip through.
If we missed something, the waitlist form below is the fastest way to ask.
An LLM rates every transcribed word from 0 (clean) to 1 (extreme). The MPAA-style rating cards (G / PG / PG-13 / R / NC-17) map directly to threshold bands. You can also fine-tune with the slider, and manually add or skip any word with a click. Default settings catch the obvious stuff; you decide the rest.
Transcription is multilingual: Whisper auto-detects the language. Bad-word rating works on the transcribed text, so any language with profanity in the LLM's training set is covered. English, Spanish, French, German, Portuguese are all well-supported. Niche languages may need manual review.
End-to-end (upload through final clean track) is typically 20–35 seconds for a 3–4 minute song. The bottleneck is upload speed, not processing.
Yes. The original upload stays in storage. Every "Finalize" call processes from the original, never from a previous shoosh. Switch from Beep to Scratch to Pitch Down freely, each gives a clean result against the original.
MP3, WAV, FLAC, M4A on upload. Output is MP3 at the same sample rate as your input.
You're responsible for having the rights to upload and edit the audio you submit. SHOOSH doesn't grant you any new rights to the underlying composition or recording; it only edits the file you provide.
Yes, once we're out of invite-only beta. Right now access is gated. Join the waitlist below and we'll get back to you.
Built by people who actually make and ship music.
Co-Founder & Production
BK/LA-based producer with 15+ years of credits: Sean Price, Roc Marciano, Griselda, CRIMEAPPLE, Freeway, Conway the Machine and many more. Co-founder of Different Worlds Music Group. Resident on WNYU's Halftime Radio Show.
Co-Founder & Music Direction
New Orleans–born, Minneapolis-based DJ and crate-digger. Co-founder of Different Worlds Music Group with Skizz. Eight-year resident host of Quiet Village on KRSM Radio. Stages shared: Durand Jones, Bun B.
Co-Founder & Engineering
Builds the AI pipeline, the infrastructure, and most of what you see on this page. Long-time audio nerd, longer-time software engineer. Says he needs his analogue compressor, eqs, deessers and gates for work meetings.
SHOOSH is invite-only while we ramp up. Drop your email and tell us what you'd use it for, and we'll be in touch as we open access.
