How to Remove Background Noise from Audio (Hum, Hiss, Room Tone)

Bad audio ruins otherwise-good recordings. A podcast with AC hum sounds amateur regardless of the content quality. A voice memo with traffic noise is hard to listen to. A Zoom call recording has so much room tone that the words are buried.

Most of these problems can be cleaned up with a single FFT-based denoising pass — the same technique professional editors use, just packaged as a one-click preset.

How to denoise audio

Audio Noise Reduction on Dropvert handles the common cases:

1. Drop the audio (or video — the tool extracts audio automatically).
2. Pick a strength: Light, Medium, or Heavy.
3. Click "Reduce noise."
4. Preview the result with the inline player. Download if you like it.

The denoising runs in your browser via FFmpeg.wasm — no upload, no signup. The output is MP3 at 192 kbps with the noise removed.

What gets removed

The denoiser targets broad-spectrum noise — sounds that are present continuously across the audio, like:

• AC hum (60 Hz / 50 Hz buzzing from electrical equipment)
• Room tone (the constant ambient noise of any indoor space — fans, distant traffic, HVAC)
• Tape hiss (continuous high-frequency noise from analog sources)
• Microphone self-noise (the constant low-level noise of cheap microphones)
• Camera fan noise (from camcorder or DSLR built-in mics)

These all share a common property: they're present across the whole recording at roughly the same level. The FFT denoiser estimates this noise floor, then subtracts it from the audio.

What doesn't get removed

The denoiser can't help with transient noise — short, distinct sounds that aren't continuously present:

• Door slams
• Coughs, sneezes
• Single clicks, pops
• Phone rings
• Loud single keystrokes

These need to be removed manually (silence the offending region in a DAW) or with specialized "click and pop removal" tools.

Three strength levels

The trade-off in noise reduction is how aggressively you remove the noise vs. how much you affect the rest of the audio. Pull too hard and the voice starts to sound underwater or robotic.

Strength	What it does	Best for
Light	8 dB reduction. Subtle clean-up, preserves max detail.	Already-good recordings with minor noise. Music.
Medium	12 dB reduction. Default, balanced.	Most podcasts and voice memos.
Heavy	20 dB reduction. Aggressive removal.	Recordings with significant hum or hiss. May affect timbre.

When in doubt, start with Medium. If the result still has noise, try Heavy. If the result sounds underwater, try Light.

How does this compare to professional tools?

Real talk: Adobe Audition's "Adaptive Noise Reduction" and iZotope RX's "Spectral De-noise" are more sophisticated than FFmpeg's afftdn (the underlying algorithm Dropvert uses). They have:

• Longer time windows (better noise floor estimation)
• AI-trained noise profiles (smarter about what's "noise" vs. "signal")
• Advanced spectral subtraction with masking thresholds
• Custom noise profile sampling (point at a quiet section, the tool learns)

For professional podcast / music / film production, those tools are still better. For "clean up my Zoom recording so it sounds less terrible," FFmpeg's afftdn is a clean free alternative.

If you need professional-grade denoising and want to stay free, Stems Separation's vocals stem can be a backdoor: separate the song into stems, use the vocals stem alone, mix back in just the music stems you want. The AI separation is more aggressive than spectral denoising but treats the vocals as a target rather than as "everything except noise."

Combine with other audio tools

After denoising, you might want to:

• Trim silence — remove dead air now that the audio is clean.
• Apply voice effects — pitch shift, robot, phone, etc.
• Transcribe — denoised audio transcribes more accurately.
• Convert format — change to a different audio format.

Common questions

Will it work on music? Light strength on music is usually safe. Medium and Heavy are aggressive enough that they can affect cymbal sparkle and high-frequency content in a noticeable way. For music, Light or skip.

Why does the output sound underwater on Heavy? Heavy denoising removes 20 dB of noise across the spectrum. If the noise floor wasn't actually 20 dB below the signal, the algorithm starts removing parts of the actual voice — high frequencies in particular. Switch to Medium.

Can I sample a noise-only section? Not in v1. The current tool uses a generic noise floor estimate. Custom noise profile (point at a quiet section, learn the noise from there) is roadmap-tracked.

Are my audio files uploaded? No. The denoising runs entirely in your browser via FFmpeg.wasm.