Reduce Audio Background Noise
Pick an audio file, set a strength, and clean out steady hiss/hum in your browser — nothing is uploaded.
About this tool
Reduce Audio Background Noise strips steady background hiss and hum — tape hiss, air-conditioner drone, mic self-noise, mains buzz — out of a recording so speech and music sit forward. Pick an audio file, set how hard to push the reduction, choose a denoiser, and download the cleaned track. Everything runs locally in your browser with a WebAssembly build of ffmpeg: your audio is never uploaded to a server.
It works best on steady, constant noise (a consistent hiss or hum that runs through the whole clip). It is not an AI speech re-synthesizer — it won't remove one-off sounds like a door slam or reconstruct a badly clipped voice.
Controls
- Reduction strength (1–100) — how aggressively to cut noise. Higher removes more but can make the audio sound hollow or "underwater." Start around 12 and raise gradually until the hiss is gone but the voice still sounds natural. Default: 12.
- Denoiser — afftdn (FFT-based, fast, the default) handles steady hiss/hum well. anlmdn (non-local means) is slower but can be gentler on transients like consonants and cymbals.
- Also cut low hum/rumble — adds an 80 Hz high-pass stage that removes mains hum, HVAC rumble, and handling thumps on top of the denoise. Default: off.
- Output format — mp3 (192 kbps, default), wav, ogg, flac, or m4a.
Worked example
Upload a noisy voice memo interview.wav with a constant air-conditioner hiss, set
strength 40, afftdn, tick Also cut low hum/rumble, and choose wav. The
tool runs afftdn=nr=38.8 behind an 80 Hz high-pass and returns
interview-denoised.wav — the same speech with the hiss and low drone knocked well
down. Prefer a smaller file? Switch the format to mp3 and you get
interview-denoised.mp3 at 192 kbps instead.
FAQ
What kind of noise can this actually remove?
It targets steady, constant background noise — a hiss, hum, or drone that runs through the whole clip (tape hiss, mic self-noise, air-conditioner or fan noise, mains buzz). It is a spectral denoiser, not an AI restorer, so it won't cleanly remove one-off events like a cough, a door slam, or street chatter, and it can't rebuild a distorted or heavily clipped voice.
What strength should I use?
Start low. 12 is a safe default that takes the edge off most hiss without touching the voice. Raise it in steps and listen: somewhere around 30–50 clears heavier noise, but push too far and the audio starts to sound hollow, watery, or "underwater" as the filter eats real signal along with the noise. If it sounds robotic, back off.
afftdn or anlmdn — which denoiser?
afftdn (the default) is an FFT/spectral denoiser: fast and good at steady broadband hiss and hum. anlmdn is a non-local-means denoiser: slower, but it can be gentler on transients (crisp consonants, cymbal hits), so try it if afftdn makes speech sound lispy or smears music. For most voice recordings afftdn is the right first choice.
What does "Also cut low hum/rumble" do?
It prepends an 80 Hz high-pass filter before the denoise, which removes low-frequency energy — 50/60 Hz mains hum, HVAC rumble, table thumps, and handling noise. Leave it off for music with real bass you want to keep; turn it on for spoken-word recordings where nothing useful lives below 80 Hz.
Is my audio uploaded anywhere? What are the limits?
No. Processing happens entirely in your browser via a WebAssembly ffmpeg build — the file never leaves your device. Practical limits: input up to about 10 MB, output up to about 10 MB. Input can be any common audio format (mp3, wav, m4a, ogg, flac); output is mp3, wav, ogg, flac, or m4a. Very long or high-bitrate files may hit the size cap — trim or compress first.
Developer & Automation Access
Run it from the terminal
Same engine as this page, headless — via the gizza CLI:
gizza tool audio-noise-reduce 'url=https://example.com/input' 'strength=12' 'method=afftdn' 'remove_hum=true' 'format=mp3'New to the CLI? Get gizza →
Open it by URL
Pre-fill and auto-run this tool with query parameters — the names match the API/CLI:
https://gizza.ai/tools/audio-noise-reduce/?url=https://example.com/input&strength=12&method=afftdn&remove_hum=true&format=mp3Machine-readable descriptor: tool.json — title + parameters JSON Schema for agents.