Why wind noise is harder than hiss or hum
Wind noise is not a single frequency. It is a low-frequency rumble, a broadband gust, and sometimes a violent pop when air hits the microphone diaphragm. That combination is why a simple noise gate or denoiser often fails. The low rumble lives under 150 Hz, but gusts can push energy into the 200–500 Hz range and make speech sound muddy. If the wind was strong enough to clip the mic preamp, you also get crackle that no EQ can rebuild. The goal is not to remove all low end; it is to separate voice from wind without making the speaker sound thin, robotic, or underwater.
Step 1: Diagnose the wind noise before you process
Put on headphones and listen to a 10-second section with speech and a gust. Then look at a spectrogram in a free editor like Audacity, Reaper, or Adobe Audition. You are checking three things:
- Frequency range: Constant rumble below 100–150 Hz? That is easy to high-pass. Gusts spread up to 300–500 Hz? You need dynamic control or AI.
- Clipping: Flat-topped waveforms mean distortion. Fix that first or every later step will exaggerate it.
- Timing: Does wind hit between words, over words, or both? Wind over speech is the hardest case.
Step 2: Fix clipping and obvious gusts first
If the waveform is clipped, use a clip repair tool. In Audacity, try Effect > Volume and Compression > Clip Fix. In iZotope RX, use De-clip. Set the threshold just below the clipped peaks and process only the damaged sections. Do not apply clip fix to the whole file if only one gust clipped. For isolated gusts that do not contain speech, select those regions and reduce them by 12–20 dB or replace them with room tone. This prevents the compressor from pumping later.
Step 3: Use a high-pass filter, but don't overdo it
A high-pass filter is the first real tool for wind rumble. In Audacity, use Effect > Filter Curve EQ or High-Pass Filter. Start at 80 Hz for a deep male voice, 100 Hz for a female voice, and 110–120 Hz for a child or higher-pitched speaker. Use a slope of 12 dB per octave, then increase to 24 dB per octave only if rumble remains. Listen for tradeoffs: too high and the voice loses chest and warmth; too steep and you can hear a resonant thump at the cutoff. If the wind still sounds like a low roar, move to dynamic processing.
Step 4: Tame gusts with dynamic EQ or multiband compression
Static EQ cuts low end all the time, including when there is no wind. A multiband compressor or dynamic EQ is smarter. Target the 20–250 Hz band. Set attack fast, around 1–5 ms, so it catches the gust. Set release around 50–150 ms so it does not pump. Use a ratio of 4:1 to 8:1 and a threshold around -30 to -20 dB, then lower it until gusts stop jumping out. If your editor has spectral editing, select a wind-only region and attenuate it directly. Tradeoff: aggressive low-band compression can dull plosives like P and B, and it can make the voice sound small. Use the minimum amount that works.
Step 5: Use AI voice isolation for specific background noises
When wind overlaps speech, EQ and compression cannot fully separate them. This is where AI voice isolation for specific background noises helps. Instead of removing all noise, it identifies voice and suppresses targeted sounds like wind, traffic, or crowd. A free online option like ClearVoice can do this without installing a DAW. Upload your file, choose the wind or outdoor noise profile if available, and run the process. For a browser-based option, try ClearVoiceHQ (clearvoicehq.com). Start with a medium strength. If you go too high, the voice can sound robotic or lose consonants. If the wind is only in a few spots, process those sections separately and keep the clean parts untouched.
Step 6: Clean up the AI result with light EQ and leveling
After AI isolation, the voice may sound dull or harsh. Use a gentle EQ move rather than a big curve. If it sounds muddy, cut 200–400 Hz by 1–3 dB. If it sounds thin, do not add back low end blindly; add a small shelf at 100–150 Hz only if the voice needs weight. If consonants sting, cut 1–3 kHz by 1–2 dB. Then normalize peaks to -1 dBTP and set loudness to -16 LUFS for podcasts or -14 LUFS for YouTube. For voice memos and Zoom recordings, -16 LUFS is a safe target. Use a limiter only for safety, not to crush the voice.
Step 7: A/B test on real devices before exporting
Do not trust one pair of studio headphones. Listen to a bypass and processed version on phone speakers, laptop speakers, and earbuds. Phone speakers reveal whether the voice is thin. Earbuds reveal artifacts like underwater warble. If the AI result sounds unnatural, blend it with the original. In a DAW, duplicate the track, process one copy, and mix the processed copy at 70–80% with the original at 20–30%. Check for phase issues and nudge if needed. If you cannot blend, reprocess at a lower strength.
How to apply this to common recording types
Phone voice memo
Export the memo as WAV or M4A, import to Audacity, and follow steps 2–6. Phone mics are tiny, so wind often clips. If the memo is clipped, repair before AI isolation.
Zoom call
Download the local or cloud recording, extract the audio, and process the speaker track. Zoom background noise suppression helps with fans and keyboards, but wind is usually too dynamic. If you only have a mixed recording, AI isolation will also affect other voices. Process a copy and compare.
Podcast interview recorded outdoors
Process each microphone track separately. If you recorded a single mixed file, use AI voice isolation first, then EQ. Do not apply the same high-pass to every speaker; use 80 Hz for low voices and 100–120 Hz for higher voices.
Gaming clip or Discord audio
If you record separate mic and game tracks, process only the mic track. If it is mixed, AI isolation may pull down game sounds. Use a medium setting and check the mix. For clips under a minute, manual spectral editing can be faster than a full AI pass.
Settings cheat sheet
- High-pass: 80–120 Hz, 12–24 dB/oct.
- Low-band compressor: 20–250 Hz, attack 1–5 ms, release 50–150 ms, ratio 4:1–8:1.
- AI voice isolation: medium strength, process in sections.
- Post EQ: cut 200–400 Hz by 1–3 dB if muddy; cut 1–3 kHz by 1–2 dB if harsh.
- Export: -1 dBTP peak, -16 LUFS podcast, -14 LUFS YouTube.
What you can't fix and how to prevent it next time
If wind caused hard clipping across syllables, no tool can fully restore the original voice. You can reduce the damage, but consonants may stay distorted. For future recordings, use a foam windscreen or a furry deadcat, enable a low-cut filter on the recorder at 80–100 Hz, and place the mic out of direct airflow. Record a backup take if possible. If you are outdoors and cannot avoid wind, record a few seconds of room tone so you can patch gaps. With the right order—repair, filter, dynamic control, AI isolation, light EQ—you can make an outdoor voice recording sound clean enough for a podcast, Zoom call, gaming clip, or voice memo.