Reduce Noise: Audio (RNNoise)

Documentation Documentation - English
1

Reduce Noise: Audio (RNNoise) was added in Shotcut version 26.06.25

RNNoise is a deep-learning-based noise suppression library.
It reduces background noise in speech recordings using the RNNoise library, a real-time speech enhancement algorithm that combines traditional digital signal processing with a Recurrent Neural Network (RNN). Its primary goal is to suppress unwanted noise while preserving the intelligibility and natural character of human speech.

image
image578Ă—241 12.4 KB

Unlike traditional noise reduction filters that require learning a noise profile or rely solely on thresholds, RNNoise has been trained to distinguish speech from many common background noises such as fans, traffic, office ambience, or air conditioning.

Parameters

Suppression (0.0–100.0%)

Controls the amount of background noise reduction applied to the audio.

  • 0.0%
    No noise suppression.

  • Low values (10%–40%)
    Mild noise reduction. Background noise is reduced while preserving most of the original ambience.

  • Medium values (40%–70%)
    Good balance between speech clarity and natural sound. This range is suitable for most voice recordings.

  • High values (70%–100%)
    Aggressive suppression. More background noise is removed, but speech may begin to sound processed or lose subtle details.

Note:

Increase Suppression only until the background noise becomes unobtrusive. Higher values do not necessarily produce better-sounding audio.

Keyframes

This filter does not support keyframes.
Noise reduction is applied continuously throughout the clip.

How RNNoise differs from traditional noise reduction

Traditional noise reduction often:

  • Learns a noise profile from a silent section.
  • Works best with constant background noise.
  • Can introduce metallic or “underwater” artifacts when pushed too far.

RNNoise instead:

  • Identifies speech and background noise using a trained neural network.
  • Adapts automatically without requiring a noise sample.
  • Generally produces more natural speech while reducing a wide variety of common noises.

Recommended use cases

  • Voice-overs
  • Interviews
  • Podcasts
  • Screen recordings
  • Online meetings
  • Field recordings with moderate background noise

RNNoise performs best when speech is the primary subject of the recording.

Limitations

  • Optimized for human speech, not music or environmental recordings.
  • Very high Suppression values may reduce speech quality or introduce audible processing artifacts.
  • Does not replace good microphone placement or proper recording conditions.
2 Likes
User Guide - Table of Contents
2

This topic was automatically closed after 2 minutes. New replies are no longer allowed.