Voice Cleaner (denoise)
Removes constant background noise and ambient sound behind speech, such as wind, insects, air conditioning, electrical hum, and keyboard clicks, while keeping the detail of the voice itself.
Loading product details
Local voice denoising and de-reverb for Mac and Windows
PureRefine handles two common problems in voice recordings: background noise (wind, air conditioning, electrical hum, keyboard clicks) and room reverb (echo in empty rooms and meeting rooms). Pick a file, check denoise, de-reverb, or both, and it produces a new audio file while the original stays untouched. The AI model runs on your computer; it downloads once on first use and works offline after that.
Removes constant background noise and ambient sound behind speech, such as wind, insects, air conditioning, electrical hum, and keyboard clicks, while keeping the detail of the voice itself.
Reduces echo from empty rooms and untreated meeting rooms, making the voice sound closer to the microphone.
Both effects happen in a single processing pass, so the model runs only once. Faster than denoising, exporting, then de-reverbing, and one less re-encode.
Process one recording or pick a folder for batch processing. Supports WAV, MP3, M4A, AAC, FLAC, AIFF, CAF, and OGG.
The AI model runs on your computer and downloads once on first use; after that no network is needed. Recordings and results are never uploaded.
Process a file or folder with a single terminal command, with input, output, and mode options. Useful for automatic cleanup after recording.
There is no equalizer or threshold to tune. If the result isn’t what you want, try another mode.
| Mode | What it handles | Typical cases |
|---|---|---|
| Denoise | Constant background noise and ambient sound | Outdoor recordings, air-conditioned rooms, computer fans |
| De-reverb | Echo and tail from room reflections | Empty rooms, meeting rooms, bathroom-like echo |
| Denoise + de-reverb | Both at once, single model pass | Voice recorded in a loud, empty space |
PureRefine is built for speech. Music and instrument recordings are outside its design and may lose tone. If a recording is badly distorted, clipped, or missing frequency ranges, repair it with PureRestore first, then polish it with PureRefine.
Both platforms are paid downloads, buy once, with no in-app purchases or subscriptions. Folder batch processing and the command line are included after purchase, and the processing quality and supported formats are the same. The difference is how they compute:
| Mac | Windows | |
|---|---|---|
| Compute | Apple Silicon | CPU |
| Model download | Via the App Store resource download service | From Efficient Lab’s model server |
| CLI file access | Only directories authorized in the app | Direct access to the paths you give |
The Windows version runs on the CPU, so no dedicated GPU is needed, but long recordings take a while. Both versions currently have an English-only interface.
The CLI uses the same processing pipeline as the interface. Handy for running a cleanup pass right after recording a podcast:
clean --input <file-or-folder> --output <output> --mode voice-cleaner | reverb-remover | both
When the input is a folder, every supported audio file inside is processed in turn. The Mac version runs in a sandbox, so the CLI can only reach directories authorized in the app; the Windows version accesses the paths you give directly.
Denoise and de-reverb both compute on your computer; recordings and results are never uploaded. PureRefine has no account system and no usage statistics. It goes online in only one case:
No. Processing happens on your computer. The network is only used for the first model download; it works offline after that.
Yes. Choose "denoise + de-reverb" and both effects finish in one pass, with the model running only once. No need to export twice.
Both the Mac and Windows versions are paid downloads, buy once. After purchase everything is included, including folder batch processing and the command line. No in-app purchases and no subscription.
Not recommended. The model is trained for speech; on music or instrument recordings it may lose tone. To separate vocals from accompaniment, use PureSpleeter.
PureRefine removes noise and echo, and suits voice recordings that are mostly intact. PureRestore handles more damaged audio, such as missing frequency ranges or heavy distortion. You can use them in sequence: repair first, then polish.
It needs macOS 14 or later and a Mac with Apple Silicon (M series). Intel Macs cannot run it.
The result is saved as a new audio file and the original is never overwritten. On Windows it lands next to the source with `_pure_refine` appended to the name; on Mac the suffix is `-pure-refine`.
PureRefine is designed for offline local processing and does not depend on cloud transcoding. Read ourprivacy policyfor full data-handling details.
You can reach us via email [email protected] or submit issues and suggestions through GitHub Issues.
Local music and sound-effect asset manager for Mac and Windows: AI tagging, batch organizing, safe writeback
Local stem separation for Mac and Windows: split a song into vocals, accompaniment, drums, bass, and more
Local audio restoration for Mac and Windows: repair recordings with hiss, noise, and missing frequency ranges
Batch-extract a whole folder of SACD ISOs into tagged DSF files