Why mouth noise is stubborn in Audacity
Mouth noise is not one problem. It is a mix of clicks, lip smacks, saliva pops, tongue clicks, and breaths. In Audacity, these noises are hard to remove because they are short transients, not steady hiss. Noise Reduction is built for constant background noise like a fan or hum. Use it on mouth clicks and you will get weird artifacts. Manual editing uses Repair, Silence, Amplify, and Spectral Delete to fix each spot. AI voice isolation takes a different approach: it separates voice from everything else, which can catch many mouth noises in one pass but can also soften consonants if pushed too hard.
Step 1: Prep your Audacity session
- Duplicate the track first. Select the track and press Ctrl+D. Mute or delete the original later. This gives you a safety net.
- Switch the track view from Waveform to Spectrogram. Click the track name dropdown and choose Spectrogram. Mouth clicks are much easier to see than hear.
- Use headphones. Laptop speakers hide clicks and smacks in the 2-8 kHz range.
- Zoom in with Ctrl+1. Fit the whole project with View > Fit Project in Window. Work in 2-5 second chunks.
- Add a Label Track: Tracks > Add New > Label Track. Press Ctrl+B to mark each mouth noise as you find it.
Step 2: Find the mouth noise visually and audibly
In Spectrogram view, different mouth noises look different. A sharp mouth click looks like a thin vertical line that runs across many frequencies, often 2-8 kHz. A lip smack looks like a low-mid blob around 500 Hz to 3 kHz. A breath looks like a fuzzy broadband patch. A saliva pop can look like a small dark spot with a bright edge.
Play 2-3 seconds around each suspicious spot. If it sounds like a click but you cannot see it, zoom in further. Use Ctrl+B to label it. Label everything first instead of fixing in real time. This keeps you from losing your place and makes the edit pass faster.
Step 3: Remove isolated clicks with Repair
For a single click, zoom in until you can see the spike clearly. Select only the click plus 2-5 milliseconds on each side. Do not select the whole word. Then go to Effect > Repair. Repair works best on very short selections, usually under 128 samples. If the selection is too long, Audacity may disable the effect or create a dull, smeared sound. Undo with Ctrl+Z if the word loses crispness. For several clicks in a row, repair each one individually. It is slower, but it sounds natural.
Step 4: Silence or attenuate lip smacks and saliva pops
Lip smacks are longer than clicks, so Repair is the wrong tool. Use one of these manual methods:
- Zoom in on the smack. Select from just before the noise to just after it. Keep the selection tight so you do not clip the surrounding word.
- If the smack sits in a pause, use Edit > Remove Special > Silence Audio (Ctrl+L). This replaces the selection with digital silence. It is fast, but it can sound unnatural if the rest of the pause has room tone.
- For a more natural result, use Effect > Amplify and set it to -12 to -20 dB. This lowers the smack without creating dead air. You can also use Effect > Fade In and Fade Out on 5-10 ms at the edges.
- If the smack overlaps a word, try spectral editing. Stay in Spectrogram view, select the smack's frequency band and time range with the Spectral Selection tool, then use Edit > Remove Special > Spectral Delete. This removes only the noisy frequencies. It can leave a small hole, so follow with tiny fades.
Step 5: Use Click Removal carefully
Effect > Click Removal can catch many small clicks in one pass, but it was designed for vinyl records, not voice. Start with Threshold around 100-200 and Spike Width around 10-20. Click Preview. If the voice sounds lispy, watery, or swallows consonants, undo. For important lines, manual Repair is safer. Use Click Removal only on sections with lots of tiny clicks that would take forever to fix by hand.
Step 6: Reduce mouth rumble and breaths
Low-frequency mouth rumble and plosive pops can be reduced with a high-pass filter. Go to Effect > High-Pass Filter. Set the frequency to 80 Hz for a deep male voice, or 100-120 Hz for a female or higher voice. Use a rolloff of 12 or 24 dB per octave. Do not go above 120 Hz unless the voice is already thin, or it will sound like a phone call.
Breaths are best handled manually. Select each breath and use Effect > Amplify at -10 to -18 dB. If there are too many breaths, try Effect > Noise Gate. Set the threshold around -35 to -45 dB, Attack around 10 ms, Release around 100-200 ms, and Level reduction around -10 to -15 dB. Test on a full phrase. A gate that is too aggressive will pump the background noise and make the edit obvious.
Step 7: Clean up with EQ and de-essing
After manual edits, the track can still sound spitty. Mouth noise often lives between 3 kHz and 6 kHz. Open Effect > Filter Curve EQ and make a wide cut of 2-4 dB in that range if needed. Keep it gentle. For harsh S sounds, Audacity does not have a built-in de-esser, but you can manually select the worst S sounds and use Amplify at -4 to -8 dB. A third-party VST like TDR Nova can also work. The goal is to reduce the edge, not to remove all high frequencies.
AI voice isolation vs manual Audacity editing
Manual editing is precise. You decide exactly what gets removed, which matters for a podcast, client video, or voice-over. The downside is time. A 30-minute interview with frequent mouth clicks can take an hour or more to clean properly. AI voice isolation is faster. Tools like ClearVoice can separate voice from background noise and reduce many mouth noises in one pass. The tradeoff is control. Aggressive AI can make the voice sound underwater, soften plosives, or create metallic artifacts. It may also remove breaths completely, which can sound unnatural.
The best workflow is usually hybrid. Run a free tool like ClearVoiceHQ (clearvoicehq.com) to isolate the voice and cut steady noise, then open the result in Audacity. Use Repair for remaining clicks and Amplify for smacks. This gets you most of the way in minutes and lets you spend manual time only on problem spots.
Step 8: Export without ruining the original
Before export, listen in headphones and on a phone speaker. Check the start, middle, and end. If you used Noise Gate or EQ, bypass it quickly to compare. When you are happy, use File > Export > Export as WAV or MP3. Keep the original recording untouched. For video work, export WAV at 48 kHz and 24-bit. For voice memos or quick sharing, MP3 at 192-320 kbps is fine.
Quick Audacity settings cheat sheet
- Isolated click: select 2-5 ms, Effect > Repair.
- Lip smack in pause: select, Ctrl+L or Amplify -12 to -20 dB.
- Spectral click: Spectrogram view, Spectral Delete, 5-10 ms fades.
- Low rumble: High-Pass Filter at 80-120 Hz, 12-24 dB per octave.
- Breaths: Amplify -10 to -18 dB or Noise Gate at -35 to -45 dB.
- Harsh 3-6 kHz: Filter Curve EQ, wide -2 to -4 dB cut.
- Many tiny clicks: Click Removal, Threshold 100-200, Spike Width 10-20, preview.
Bottom line
Audacity can remove mouth noise, but it rewards patience. Use Spectrogram view to see clicks, Repair for tiny spikes, Silence or Amplify for smacks, and EQ for spitty high frequencies. If the file is long or you need a fast cleanup, let AI do the heavy lifting first. ClearVoice-style isolation plus Audacity touch-ups is often the fastest route to clean, natural voice audio.