How to Remove Hiss from Audio in Audacity: Manual Fixes vs AI Voice Isolation

Why Hiss Happens and What You Are Actually Removing

Hiss is broadband high-frequency noise: the steady shhh from a cheap preamp, laptop fan, phone mic, or bad gain staging. It is not hum at 50/60 Hz, a ground loop, or room reverb. That matters because Audacity has different tools for each. If you attack hiss with a hum notch, you will barely touch it. If you use heavy Noise Reduction, voices can sound lispy, metallic, or underwater.

Fix what you can at the source first: lower input gain so peaks hit around -12 dB to -6 dB, move the mic closer, turn off fans, and use a quiet room. If the recording already exists, Audacity can still help.

Step 1: Make a Clean Noise Profile in Audacity

1. Open your file. Save a backup copy first.
2. Find 0.5 to 2 seconds of pure hiss with no breathing, speech, or clicks.
3. Select that section, then go to Effect > Noise Removal and Repair > Noise Reduction. Click Get Noise Profile.
4. Select the audio you want to clean, or press Ctrl+A.
5. Open Noise Reduction again and use the settings below.

Audacity Noise Reduction settings that usually work

• Noise reduction: 6-12 dB. Start at 6-8 dB for mild hiss, 10-12 dB for a noisy phone memo. Avoid 18 dB or more unless the recording is unusable.
• Sensitivity: 3-5. Higher values catch more noise but can eat soft consonants.
• Frequency smoothing: 3-6 bands. Use 3 for speech, 6 if the hiss is very broad.

Preview before applying. If the voice sounds watery or robotic, undo and lower the reduction or sensitivity. Two gentle passes of 6 dB often sound better than one 14 dB pass. For a podcast or voiceover, 6-10 dB is usually the sweet spot.

Step 2: Gate the Gaps, Not the Words

Noise Reduction lowers hiss everywhere. A noise gate mutes or reduces the quiet parts between words. In Audacity, go to Effect > Noise Removal and Repair > Noise Gate.

• Gate threshold: just above the noise floor. Try -45 to -35 dB for a noisy phone recording, -55 to -45 dB for a decent mic.
• Level reduction: -6 to -12 dB. Do not use -100 dB unless you want unnatural dead silence.
• Attack: 10-20 ms. Hold: 50-100 ms. Decay: 100-200 ms. Too fast clips word starts; too long makes breaths pop.

Gates can chop reverb tails or create a pumping effect. If your recording has music, laughter, or multiple speakers, use a gentle gate or skip it.

Step 3: EQ and Filtering Without Dulling the Voice

Hiss usually lives from 4 kHz to 12 kHz, but intelligibility also lives around 2-5 kHz. Do not simply low-pass at 6 kHz or the voice will sound like a telephone. Use these moves instead:

1. High-pass filter: Effect > EQ and Filters > High-Pass Filter. Set 80 Hz for deep male voices, 100-120 Hz for most other voices. Use a 12 dB/oct rolloff. This removes rumble, not hiss.
2. Low-pass filter: start at 12 kHz, then try 10 kHz if the hiss is severe. Stay at 8 kHz or above for natural speech.
3. Filter Curve EQ: add a gentle shelf from 5 kHz to 12 kHz of -3 to -6 dB. Keep it wide. Narrow notches can make the voice sound hollow.
4. Notch filters: use only for electrical hum at 50/60 Hz and its harmonics. They will not remove broadband hiss.

If you have one short hiss burst or whine, open the track in Spectrogram view and use Spectral Delete if your Audacity version has it. Repair is better for clicks.

The Manual Workflow I Would Use on a Real Recording

1. Back up the original. Save as an Audacity project.
2. Do not normalize first. Normalizing raises the noise floor before you remove it.
3. Get a noise profile from pure hiss.
4. Apply Noise Reduction at 8 dB, sensitivity 4, smoothing 3.
5. If hiss remains, apply a second pass at 4-6 dB. If the voice sounds lispy, undo and stop.
6. High-pass at 90 Hz. Low-pass at 11 kHz.
7. Noise Gate at threshold -45 dB, level reduction -8 dB, attack 10 ms, hold 80 ms, decay 150 ms.
8. If S sounds are harsh, use Filter Curve EQ to reduce 6-8 kHz by 2-3 dB.
9. If needed, compress with Threshold -18 dB, ratio 2:1, attack 10 ms, release 100 ms. Put the gate before compression so compression does not raise the hiss.
10. Use Loudness Normalization to -16 LUFS for podcasts or -14 LUFS for YouTube. Export a WAV master, then MP3 at 192 kbps or higher.

This chain takes 10-30 minutes per file once you know it. The hard part is repeatability on long Zoom calls or changing noise.

Where Manual Editing Falls Apart

Audacity Noise Reduction is static. It treats the selected audio the same way throughout. If the hiss changes when someone moves, a fan kicks on, or a laptop shifts, one noise profile cannot handle it perfectly. You will either leave some hiss or chew up the voice. And if hiss overlaps speech, Audacity cannot truly separate voice from noise. It can only subtract a fingerprint and hope for the best.

AI Voice Isolation vs Manual Audacity Methods

AI voice isolation uses a model trained on speech and noise to predict which parts of the sound are voice. It can remove consistent hiss, fans, keyboards, and room noise while preserving more speech. It is faster for long recordings and changing noise, but it can add artifacts: watery tone, dropped consonants, or a processed sheen, especially with music or multiple speakers.

Manual Audacity is better when you want full control, the hiss is constant, or you have time to fine-tune. AI is better when you need a clean voice track fast. A hybrid approach often wins: run AI isolation first, then use Audacity for high-pass, gentle EQ, and a light gate.

If you would rather not dial in Noise Reduction settings, a free tool like ClearVoiceHQ (clearvoicehq.com) can isolate voice from hiss and background noise online. Upload your file, let it process, and download the voice track. It will not replace Audacity for surgical edits, but it is a fast first pass for a noisy voice memo, Zoom call, or gaming clip. ClearVoice is also useful for several files when you do not want to repeat the same manual chain.

Quick Settings Cheat Sheet

• Mild laptop mic hiss: Noise Reduction 6 dB, Sensitivity 3, Smoothing 3; High-pass 100 Hz; Low-pass 12 kHz.
• Phone voice memo: Noise Reduction 10 dB, Sensitivity 4, Smoothing 6; Gate threshold -40 dB, Level reduction -8 dB.
• Zoom call with fan: AI voice isolation first, then High-pass 90 Hz and a wide -3 dB shelf from 6-10 kHz.
• Gaming clip with keyboard: Gate threshold -38 dB, attack 5 ms, hold 80 ms. Avoid heavy Noise Reduction because key clicks are transient.

Final Checklist

Keep the original file. Preview every effect. Use the least processing that solves the problem. If the voice sounds worse after Noise Reduction, undo and try AI isolation or re-record with better gain.