What types of noise can it remove?

RNNoise is effective against stationary noise (fan hum, HVAC, electrical hiss) and some non-stationary noise (keyboard clicks, background conversations). It is not effective against music playing in the background or impulsive noise like door slams.

How long does processing take?

Approximately 1–3× the audio duration depending on your device. A 10-minute recording typically takes 10–30 minutes to process.

Does this work on music?

This tool is optimised for speech. Using it on music will degrade the audio quality. Use a music-specific denoiser for music recordings.

Is my audio uploaded to a server?

No. All processing runs in your browser using WebAssembly. Your audio never leaves your device.

← Back to Tools

🎙️

🌐 Runs in Browser⏱ Est. 1–3× audio durationNEW

Eliminador de Ruido de Audio

Name: Eliminador de Ruido de Audio
Author: Zeraku

Elimina el ruido de fondo de grabaciones antes de transcribir

Try It Now — Free

What is Eliminador de Ruido de Audio?

Clean up noisy meeting recordings, interviews, and voice memos directly in your browser. Removes background noise, fan hum, keyboard clicks, and room echo to improve audio quality before transcription. Entirely browser-based using RNNoise WebAssembly — your audio files never leave your device.

Key Features

RNNoise neural network denoising — effective on all types of background noise

Removes fan hum, HVAC, keyboard clicks, and room echo

Before/after waveform comparison

Adjustable noise reduction strength

Supports MP3, WAV, M4A (up to 100 MB)

100% browser-based — audio never leaves your device

One-click "Send to Transcription" workflow

How It Works

Upload Audio

Drop your MP3, WAV, or M4A recording onto the page. Supports files up to 100 MB.

Preview

Listen to the original and view the waveform. Identify noisy sections.

Denoise

Click Denoise. RNNoise processes the audio in a Web Worker using a trained recurrent neural network to separate voice from noise.

Export or Transcribe

Download the clean audio as WAV or MP3, or send it directly to the Audio Transcription tool for improved accuracy.

Who Is This For?

▸Improving transcription accuracy for noisy meeting recordings
▸Cleaning up podcast recordings with background hum
▸Processing interview audio recorded in public spaces
▸Preparing voice memos recorded on a phone for sharing
▸Improving audio quality of online lecture recordings

Why Use Eliminador de Ruido de Audio?

Adobe Podcast's Enhance Speech and Krisp.ai produce excellent results but require account creation and upload your audio to remote servers — a concern for confidential meeting recordings. Zeraku's Audio Denoiser runs the RNNoise model entirely in your browser. Your audio stays private. No subscription required. It's also the ideal first step before using Zeraku's Audio Transcription tool: cleaner audio means higher transcription accuracy.

Beginner's Guide

Background noise is any unwanted sound in a recording — fan humming, air conditioning, keyboard clicks, street traffic, or room reverberation. This noise makes recordings harder to understand and reduces the accuracy of automatic transcription. This tool uses a neural network (AI model trained on audio) to identify which parts of the sound are voice and which are noise, then reduces the noise while preserving the voice. Upload your recording, click Denoise, and compare the before and after. If you plan to transcribe the audio afterwards, use the "Send to Transcription" button to continue in the Audio Transcription tool.

Technical Details

Uses rnnoise-wasm, a WebAssembly port of Mozilla's RNNoise library — a recurrent neural network trained on speech and noise samples to perform noise suppression at the frame level (10 ms frames at 48 kHz). Audio is decoded to PCM via the Web Audio API and resampled to 48 kHz mono before processing. Frames are passed through the RNN, which outputs a gain mask; gain values below the threshold are attenuated. Processed frames are re-encoded to WAV or MP3 (via lamejs WASM) in a dedicated Web Worker. No audio data is ever transmitted to a server.