How to Remove Background Noise from Audio: 5 Free Ways

Why Noise Happens

Background noise (fan hum, hiss, air conditioning, room tone) comes from recording environments, not from your voice. A cheap microphone and an untreated room guarantee noise in every take.

Every microphone picks up whatever is happening around it: the desk fan, the fridge, traffic outside, the faint hiss of your audio interface. Turning up the gain also turns up the room. That constant layer is the noise floor. It decides whether a recording sounds professional or like a voice memo.

Method 1: AI Noise Removal (Best)

AI noise removal uses a neural network trained on thousands of noisy and clean audio pairs. It removes steady noise while keeping your voice natural. Far better than a simple filter, because it separates noise from speech instead of just cutting frequencies.

Soundtool's AI Noise Remover runs a neural network called DeepFilterNet (model file deepfilter.onnx) in your browser. First use downloads the model (about 9 MB) then everything runs locally. The network decides what is noise and what is your voice, and keeps only the voice.

• Handles fan hum, air conditioning, hiss, room tone, traffic rumble, keyboard clicks.
• The voice stays natural at default strength: raise it only for very noisy recordings.

Method 2: EQ Cutting

Hum lives at low frequencies (50-60 Hz) and hiss at high frequencies (above 8 kHz). A surgical EQ can cut these bands, but it also dulls your voice. A blunt tool.

Why 50-60 Hz? That is the frequency of mains electricity, and anything plugged into the wall can leak it into your recording. Our free Noise Reducer does exactly this: a high-pass filter cuts the low rumble, a low-pass filter cuts the hiss, and a strength slider controls the depth.

Method 3: Noise Gate

A gate mutes the audio between phrases. It hides noise in the gaps but does nothing while you are speaking. The hum is still behind every word.

A gate works on a threshold: when the signal drops below it, the audio is silenced; when you speak again, it opens back up. It shines in interviews and podcasts, where clean pauses matter more than the noise under the words. Pair it with AI removal. Remove first, then gate.

Method 4: Free Software (Audacity)

Audacity is a free, open-source audio editor for Windows, macOS and Linux, and its Noise Reduction effect is still one of the most reliable free options. The clever part: it learns your specific noise first.

Find a second or two of pure noise in your recording. A pause where nobody speaks. Select it, open Effect → Noise Reduction and click Get Noise Profile. Then select the whole track, open the effect again, and apply.

• Select a noise-only section and get the noise profile.
• Select the full track and apply Noise Reduction with moderate settings.
• Listen through once: undo and lower the strength if the voice sounds metallic.

Method 5: Fix It at the Source

The cheapest noise removal is not recording the noise at all. Get the mic close (10 to 20 cm from your mouth) so your voice dominates the room. Use a directional cardioid mic and point its back at the noise source.

• Turn off fans, air conditioning and anything with a motor while recording.
• Add a pop filter for plosives and a foam windshield to soften room reflections.
• Record in a soft room (carpet, curtains, cushions) not a bare tiled kitchen.
• Keep input gain moderate: louder just raises the noise floor.

Voice vs Music: Different Problems

Speech and music need different treatment. A voice has natural pauses and a narrow frequency range, so AI removal plus a gentle gate works beautifully. Music has no pauses to exploit, and its detail spans the whole spectrum.

Push heavy noise removal onto a song and you lose what makes it feel alive. Reverb tails, harmonics, transient snap. It turns muddy. For music, use a light touch: mild EQ cuts or very low AI strength.

When Noise Cannot Be Saved

Noise removal works when your voice sits clearly above the noise. Roughly 20 dB of distance or more. When the noise is nearly as loud as the speech, no tool can separate them cleanly: the result sounds underwater, robotic or warbly.

If a recording is that bad, re-record if you can. Closer to the mic, in a quieter room. A two-minute redo beats half an hour of cleanup.

The Free, No-Upload Way

Soundtool's AI Noise Remover runs entirely in your browser. Upload nothing. The model downloads once, and every second of audio is processed on your device. Free, unlimited, private.

Drop in a file, wait a moment for the one-time ~9 MB model download, set the strength, and process. No account, no watermarks, no file-size limits. Once the model is cached, the tool works offline.

• Open the AI Noise Remover and load your file.
• Let the DeepFilterNet model download once (~9 MB), then adjust strength.
• Process and download: your audio never left your device.

FAQ