Clean the audio. Keep the take.
Separate speech from music and noise exactly, then enhance the speech when you need only the voice. Your recording stays on your computer.
Apple Silicon Mac · Windows x64 · Linux AppImage (experimental)
· No account or credit card.
From your source to your edit.
- 01
Choose an audio or video file. SmoothyApp separates it into Speech and Music, with noise kept in Music
- 02
Turn on Enhanced to clean the speech and remove music and noise; the first use sets up the local engine
- 03
Compare the exact split with the enhanced voice and save the WAVs you need
Clean the take, not the performance
A room hum or a music bed can sit under an otherwise usable interview. Separate the voice from everything else, then choose the exact split or the enhanced speech for the edit. Keep the original file so you can return to it.
A few things to know
- The first separation downloads the engine and both models (several hundred MB) and needs Python 3.12 or uv
- Original plays the exact Speech + Music split unchanged; Enhanced plays only the cleaned speech
- Processing runs on the CPU; long recordings can take several minutes
- Separation can leave audible artifacts, so review the result before replacing anything
- Audio Cleanup saves WAV files; import them into Premiere Pro, DaVinci Resolve or another editor yourself
- Only the first audio stream in a file is used. The source is decoded to stereo 44.1 kHz before processing, and each result is saved as WAV
A few useful answers.
What does Audio Cleanup do?
Choose an audio or video file. SmoothyApp splits it into Speech and Music, with noise kept in the music. Turn on Enhanced to clean the speech and remove music and noise completely, then use Less / More on the enhanced speech preview and save the WAVs you need.
Is Audio Cleanup free?
Yes. It runs on your computer and uses no SmoothyEdit account, Studio plan or cloud credits. Only the one-time engine and model download needs internet.
What do I need to run it?
The first setup needs Python 3.12 or uv, an internet connection and several hundred megabytes for the engine and models. After that, processing works offline on the CPU. Long recordings can take several minutes.
Does it put the cleaned audio on my timeline?
No. Audio Cleanup saves the tracks as WAV files. Import the files into Premiere Pro, DaVinci Resolve or another editor and place them where you want them.
How good is the separation?
It depends on the recording. Clear dialogue with a steady music bed separates more cleanly than overlapping voices, heavy reverb or loud effects. Some artifacts can remain, so listen to the result before replacing your original.
Can I keep the removed noise?
Noise is not a separate track. It stays with the music in the exact Speech + Music split, so save the Music WAV if you want the room sound or noise.
Can I adjust the music?
The exact split is unchanged and has no level controls. Enhanced removes music and noise entirely and shows Less / More for the enhanced speech preview only; the saved WAV keeps the enhanced speech.
Which models does it use?
Voice/music separation uses a UVR MDX-Net model, and speech enhancement uses UVR DeNoise Lite. The models are downloaded from the upstream UVR catalog during setup and keep their own terms; SmoothyApp’s own source remains MIT.
Does it upload my audio?
No. The source file, decoding, model processing and WAV output all stay on your computer.
Try it with a guide.
How to Clean Up Interview Audio Locally (Noise and Music Separation)
Split an interview into speech and music, keep noise with the music, or enhance the speech and remove everything else with SmoothyApp Audio Cleanup. Free, local and saved as WAV files.
How to Make Your Edits Sound Professional
Sound design is what separates an amateur edit from a pro one, and it's mostly about timing. Here's the general approach to sound effects and music — and how to plan the cues before mixing them in your editor.
Keep the useful tools close.
Free desktop tool. No SmoothyEdit account or Studio credits. The one-time engine setup needs Python 3.12 or uv and an internet connection.
Apple Silicon Mac · Windows x64 · Linux AppImage (experimental)
· No account or credit card.