Audio lab · Noise filtering
Low-pass filter noise reduction: interactive demo
A low-pass filter improves signal-to-noise ratio when the wanted signal sits mostly below the cutoff and the noise is spread across the whole band. In this demo you set the input SNR and the cutoff of a Butterworth filter, then hear and measure the result. Set the cutoff too low and the filter removes part of the wanted signal along with the hiss.
These demos are illustrative. Every signal is synthetic, generated and processed in your browser, and every number shown is computed from those signals. They show the principles, not the measured performance of a product.
Press a listen button to play a short sample. Sound starts only when you press a button. No microphone or uploads needed.
Signal processing
Find the signal in the noise.
A repeating musical tone, mixed with broadband hiss. How much can a simple filter help?
Shape the sound
Lower values mean more noise.
Lower cutoffs remove more hiss—and can dull the wanted signal.
What passes through?
Filter responseBlue: what the filter lets throughDashed: the cutoff, set with the Low-pass cutoff slider
Output SNR compares filtered signal power with filtered noise power. Signal retained shows output signal power relative to the clean input; listen for distortion as well as noise reduction.
Sensor conditioning, speech front ends and embedded audio. Filtering is a useful baseline; separating overlapping speech and noise calls for more advanced methods.
What this experiment models
A deterministic four-second harmonic melody at 24 kHz, with white noise scaled to the chosen input SNR, processed by a second-order Butterworth low-pass filter. The clean and noise components are filtered separately to calculate the displayed SNR. Playback uses the same gain across comparisons; there is no per-sample loudness normalization. This is classical DSP, not a trained speech-enhancement model.
The principle
How it works
White noise has equal power per hertz, so most of its energy lies far above the few hundred hertz to low kilohertz where a melody’s harmonics sit. A low-pass filter attenuates everything above its cutoff. The noise power that passes shrinks with the bandwidth kept, while a signal concentrated below the cutoff passes almost unchanged.
The demo filters the clean melody and the noise separately with the same second-order Butterworth filter. That is why it can report the output SNR and the share of signal power retained exactly, instead of estimating them. Move the cutoff below the melody’s upper harmonics and the retained figure drops and the tone sounds duller: a better SNR number does not always mean a better-sounding signal.
Filtering is the baseline in sensor conditioning and audio front ends. When speech and noise overlap in frequency, a fixed filter cannot separate them, and adaptive methods such as spectral subtraction or trained models take over. For that work, see noise reduction and anc.
Audio lab
More demos
All six demos on one page: the audio labNoise reduction and ANC →