How to Remove Background Noise from Audio Online

Identify Your Noise Type First

Different noise types require different solutions. Using the wrong tool wastes time and degrades audio quality.

Electrical hum (60 Hz or 50 Hz): A characteristic buzzing or droning tone with a consistent pitch. Common sources: ground loops in audio equipment, cheap USB cables near power lines, fluorescent lighting. The hum has a fundamental frequency (60 Hz in North America, 50 Hz in Europe) and harmonics (120/240 Hz or 100/200 Hz). This is a narrow-frequency problem — an EQ notch filter solves it completely for free.

HVAC and fan noise: Steady broadband noise — a continuous, spectrally wide hiss or rumble. This covers a wide range of frequencies and cannot be addressed with a simple notch filter. AI noise removal is the effective solution.

Wind noise: Low-frequency, irregular rumble. A high-pass filter (cutting everything below 80–120 Hz) reduces wind noise substantially. AI noise removal handles remaining wind noise.

Room reverb / echo: Not technically "noise" but perceived as poor quality. The sound of a voice in a hard-walled room — a short flutter echo. Cannot be removed after recording with current technology without introducing significant artifacts. Prevention is the only reliable solution.

Keyboard, mouse, and handling noise: Transient sounds with irregular timing. These require manual editing (cutting around each occurrence) or AI noise removal if the transients are frequent. Automation is unreliable for transient noise.

Crowd / ambient noise: Complex, constantly varying noise. Requires AI; the most challenging type for any noise reduction system.

Free Method: EQ Notch Filters for Hum Removal

If your noise is electrical hum, the Equalizer tool solves it for free with no AI required.

How notch filtering works: A notch filter is a very narrow EQ band that cuts a specific frequency dramatically (−20 to −30 dB) while leaving all surrounding frequencies unaffected. A 60 Hz notch eliminates the fundamental hum frequency while keeping bass frequencies just above it intact.

Setting up hum removal in the Equalizer:

1. Open the Equalizer tool and upload your audio

2. Listen to the recording — identify whether the hum is 50 Hz (Europe/Australia/Asia) or 60 Hz (North America)

3. Add a notch band at the fundamental frequency (50 or 60 Hz):

• Q value (width): 10–20 (very narrow)
• Gain: −20 to −30 dB

4. Add a second notch at the first harmonic (100 Hz or 120 Hz)

5. If hum persists: add a notch at the second harmonic (150 Hz or 180 Hz)

6. Preview — the hum should be gone or significantly reduced

7. Check that speech is unaffected (the human voice fundamental is 100–300 Hz for men, 200–400 Hz for women — keep your notches narrow to avoid cutting speech)

When this is not enough: If the hum is very loud relative to your signal, notch filtering reduces but does not eliminate it. The hum components smear across adjacent frequencies through intermodulation, which narrow notches cannot address. AI noise removal handles these cases.

Free Method: High-Pass Filter for Rumble and Wind

A high-pass filter removes all frequencies below a cutoff point. Since human voice content is concentrated between 100 Hz and 8000 Hz, cutting everything below 80–100 Hz removes rumble, wind noise, and handling noise without touching the voice.

High-pass filter settings for voice recordings:

• Cutoff frequency: 80 Hz for a natural-sounding result; 100 Hz if more rumble reduction is needed
• Filter slope: 12 dB/octave is gentler; 24 dB/octave cuts more aggressively below the cutoff
• For podcast voice: 80 Hz with 12 dB/octave is the standard podcast EQ move — it eliminates mechanical vibrations and proximity-effect low-end buildup without thinning the voice

What this does NOT remove: Broadband noise (HVAC, fans) that exists across the full frequency spectrum. A high-pass filter only removes low-frequency content. If your noise is a wide hiss, this will not help significantly.

Stack high-pass with notch filters: For recordings with both rumble and electrical hum, apply a high-pass filter at 80 Hz and a notch at 60 or 50 Hz simultaneously. This handles both noise types in one EQ pass.

AI Noise Removal for Broadband Noise

For steady broadband noise — HVAC, fan, traffic, air conditioning — AI noise removal is the only reliable solution. The AI model analyzes your audio, learns the noise profile, and applies spectrally adaptive filtering to separate speech from noise across all frequencies simultaneously.

How AI noise removal works: Deep learning models trained on thousands of hours of noisy and clean audio learn to distinguish speech patterns from noise patterns. The model operates in the frequency domain, applying a time-varying mask that suppresses noise while preserving speech energy. Modern models (RNNoise, DeepFilterNet, and custom architectures) can achieve 15–25 dB noise suppression while preserving speech clarity.

Aggressiveness settings:

• Low (20–40%): Subtle noise reduction. Noise is reduced but still present. Good for recordings where the noise is a minor distraction and naturalness is paramount.
• Medium (50–70%): Significant reduction for most recordings. The recommended starting point. Noise is largely eliminated; some residual may remain in very quiet passages.
• High (80–100%): Maximum suppression. Noise is essentially eliminated but the risk of processing artifacts (swimming, metallic sound, speech distortion) increases significantly. Use high aggressiveness only for recordings with severe noise and where some quality loss is acceptable.

The over-processing warning: AI noise removal artifacts are often more distracting than moderate background noise. A podcast recording with a quiet fan hum sounds professional; the same recording with "swimming" artifacts from over-aggressive noise removal sounds amateur. Start at medium aggressiveness and only increase if results are insufficient.

Process order: Apply noise removal BEFORE normalization. Noise removal changes the overall level and spectral balance of the audio. Normalize after, to set the final loudness.

Recording Tips to Prevent Noise

The most effective noise reduction happens before recording. Post-processing can reduce noise; it cannot fully eliminate it without some cost to audio quality. Thirty minutes of acoustic setup before recording saves hours of editing.

Microphone placement: Keep the microphone 15–30 cm from your mouth. Closer means more voice signal relative to room noise. The signal-to-noise ratio improves dramatically with distance — halving the distance to your microphone doubles the signal level while keeping noise level constant, a 6 dB improvement.

Turn off noise sources: Air conditioning, ceiling fans, computers with loud fans. If the room has an HVAC system you cannot turn off, place the microphone to face away from the air vents.

Close doors and windows: Reduce external traffic and HVAC duct noise. Even partially closing a door reduces transmission significantly.

Soft furnishings absorb sound: Recording in a room with carpets, curtains, bookshelves, and upholstered furniture gives you less reverb than a bare-walled room. A closet full of clothes is a surprisingly effective recording booth — the fabric absorbs reflections from all sides.

USB audio and ground loops: If you have electrical hum, check your USB cable (replace with a shielded cable), try a different USB port, and ensure your audio interface is connected to the same power strip as your computer. Ground loop hum is eliminated at the source, not in post.

Test before recording: Record 30 seconds of silence in your recording environment before the session. Play it back at high volume to hear what your recording environment actually sounds like. Address anything audible before recording your actual content.