NAUPLY

Clarea tutorial: clean up a podcast or voice-over recording

2 min read · Updated

Clarea cleans up a recorded voice: background noise (with a neural network running inside the plugin), room reverb, breaths, plosives, rumble and uneven level. For podcasts, voice-overs, interviews and video voices. Insert it on the vocal track, first in the chain. The numbers match the markers on the screenshot.

Step by step

NAUPLY Clarea tutorial: clean up a podcast or voice-over recording interface, with numbered markers for each step

  1. Insert Clarea

    Insert Clarea on the vocal track, before EQ and compression, and open its window.

  2. Pick a preset

    Start from a preset close to your recording: Podcast & Voice-Over > Podcast Home Studio, Interview & Location > Street Interview, Podcast & Voice-Over > Untreated Room Voice…

  3. Switch on the modules you need

    Or switch ON only what is needed: usually 2 AI Denoise first, then 3 De-reverb if the room is audible, then 1 Rumble for low thumps.

  4. Listen to what is removed

    A module's LISTEN plays only what it takes out: if you hear words in it, lower its REDUCTION.

  5. Handle breaths, plosives and level

    Breaths too loud: 4 Breath (6 to 10 dB stays natural). Popping "p" and "b": 5 Plosive. Uneven level: 6 Leveler, set TARGET.

  6. Compare and blend

    Compare with BYPASS (same latency), then use MIX to bring back some of the original voice if it sounds over-processed.

The order of the six modules

The order is fixed: 1 Rumble, 2 AI Denoise, 3 De-reverb, 4 Breath, 5 Plosive, 6 Leveler, then MIX and OUTPUT. Each row has its ON, its LISTEN, its knobs and a live status.

Rumble is a high-pass filter: 70 to 90 Hz for a male voice, 100 to 150 Hz for a high voice or wind. De-reverb shortens the room tail (REDUCTION, TAIL, or AUTO which estimates it by itself when there are pauses).

AI Denoise: a local neural network

AI Denoise is a neural-network voice denoiser computed on your computer's processor: no internet, nothing is sent. REDUCTION limits how much noise is removed; Full lets the network remove everything it identifies as noise. The model status and network load are shown at the top of the window.

Breaths, plosives and level

Breath lowers breaths between phrases (REDUCTION, SENSITIVITY). Plosive tames "p" and "b" sounds (MAX REDUCTION, FREQ, SENSITIVITY). Leveler follows the voice to bring it toward a target (TARGET, RANGE, SPEED: slow for phrases, fast for words); it is not a compressor and does not raise the pauses.

What about singing?

The network learned mostly on speech. On a sung vocal, keep REDUCTION low and check sustained vowels by ear before committing. Clarea does not separate a voice from music: that takes a stem separation tool.

Its fixed latency, about 110 ms at 48 kHz, is compensated by the DAW: it is not a tool for monitoring yourself live on headphones.

For damaged audio

Clipping, clicks, electrical hum or noise on an instrument: Repair AI covers these repairs with five modules and spectral repair.

Frequently asked questions

Does Clarea send my voice over the internet?
No. The neural network runs on your computer's processor, with no connection.
Can Clarea be used live?
No: its latency of about 110 ms is compensated on playback, but it rules out live monitoring.
Does Clarea work on singing?
It learned mostly on speech. On singing, keep the reduction moderate and check sustained vowels by ear.

NAUPLY news

Plugin releases, updates and offers. A few emails a year, one-click unsubscribe.

More guides