Noise Filter LabDeepFilterLab waveforms and similarity matrix
DeepFilterNet is used under the MIT license.Copyright © 2021 Hendrik Schröter

Controlled listening test

Measure what the filter keeps.

Combine clean speech with real-world noise, process every pairing with DeepFilterNet noise reduction, and compare the effectiveness of results in one precise listening workspace.

How this works:

Clean and noisy liquids being mixed in a laboratory beaker
1. Mix speech with noise
A laboratory filter separating impurities from liquid
2. Filter out noise
A researcher comparing filtered and unfiltered samples
3. Compare results
Set up a test
01

Source material

Start with the included set, or make it yours.

Use up to five files in each group. Empty rows and columns remain visible in the results matrix.

Speech files

Voices A–E

Default speech A.mp3Included · mono · 48 kHz
Default speech B.mp3Included · mono · 48 kHz
Default speech C.mp3Included · mono · 48 kHz
Default speech D.mp3Included · mono · 48 kHz
Default speech E.mp3Included · mono · 48 kHz
5 of 5 selectedNormalized to −12 dBFS peak

Noise files

Noise 1–5

Default noise 1.mp3Included · mono · 48 kHz
Default noise 2.mp3Included · mono · 48 kHz
Default noise 3.mp3Included · mono · 48 kHz
Default noise 4.mp3Included · mono · 48 kHz
Default noise 5.mp3Included · mono · 48 kHz
5 of 5 selectedNormalized to −12 dBFS peak
25 combinations available60 seconds each · mono WAV · 48 kHz
More speechMore noise
02

Results matrix

Similarity, combination by combination.

Select a speech and noise combination to play back below.

Overall similarity score90.5%Average of individual scores
Speech files →
↓ Noise filesASpeechBSpeechCSpeechDSpeechESpeech
1Noise
2Noise
3Noise
4Noise
5Noise
More similarity More difference

Speech A with noise 1

Similarity score scale is 0–100%

The similarity score estimates how closely the noise-reduced audio matches the original clean speech by comparing the RMS level of their difference with the RMS level of the original.

Back to test setup
Listen to this combination
03

Synchronized playback

Listen across the same instant.

All four versions share one transport. Switch sources while they continue from the exact same point.

A1

Speech A with noise 1

0:00
0:60
Back to results Back to setup