If you have ever tried to clean up a voice recording, you have probably met two tools that sound like they do the same job: a noise gate and noise reduction. They don't. Picking the wrong one is the most common reason cleaned audio ends up sounding choppy, robotic or strangely empty.
Here is what each one actually does, where each fits, and how to tell which you need.
What a noise gate does
A noise gate is a volume-controlled switch. You set a threshold. When the sound is louder than the threshold, the gate is open and everything passes through. When the sound drops below it, the gate closes and the output goes quiet (or is turned down by a set amount).
That's all it does. It does not know what a voice is or what noise is. It only knows loud and quiet.
Where a gate helps:
- Turning down the hiss or fan noise in the gaps between sentences, when the noise is much quieter than the voice.
- Cutting spill from other microphones in a multi-microphone setup, so each track is only "on" when its speaker talks.
- Tidying a guitar amp or other instrument that buzzes between phrases.
Where a gate fails:
- Noise under the voice is untouched. While someone is speaking the gate is open, so the fan, the traffic or the hiss plays right along with every word.
- Word endings get clipped. Soft consonants, breaths and the tail of a word are often quieter than the threshold, so the gate shuts too early. Speech starts to sound chopped.
- The background pumps. The noise appears when someone talks and vanishes when they stop. Listeners notice that switching more than they would have noticed a steady, quiet background.
What noise reduction does
Noise reduction works on the sound itself rather than on its volume. Instead of switching the whole signal on and off, it tries to separate the wanted sound from the unwanted sound and lower only the unwanted part, including while someone is speaking.
There are two broad approaches.
Classic spectral noise reduction
The traditional method asks you to select a few seconds of background only, learns that "noise profile", and then subtracts something like it from the whole recording. This works reasonably well for steady noise such as hiss or a constant fan. It struggles with noise that changes (a passing car, a keyboard, a dog), and if you push it too hard you get the familiar watery, underwater or "musical" artefacts.
AI speech enhancement
Newer tools are trained on speech. Rather than learning one noise profile, they recognise what a human voice sounds like and keep that, while removing whatever else is around it. Because they don't rely on a fixed profile, they can deal with changing noise too, and they don't need you to find a section of pure background.
Our tool uses this approach: our speech AI keeps the voice and removes the sound around it, and because it is not a gate, it doesn't chop word endings or pump the background between sentences.
Side by side
| Noise gate | Classic noise reduction | AI speech enhancement | |
|---|---|---|---|
| Removes noise under speech | No | Partly | Yes |
| Handles changing noise | No | Poorly | Usually well |
| Needs a noise sample | No | Yes | No |
| Risk of chopped words | High | Low | Low |
| Typical artefact | Pumping, clipped tails | Watery, metallic sound | Over-processing if the voice is very weak |
| Best for | Gaps between phrases | Steady hiss and fan noise | Spoken word with real-world noise |
How to tell which you need
Play a section of your recording where someone is talking and ask one question: can I hear the noise while they speak?
- No, only in the pauses. A gentle gate, or an expander (a softer version that turns the gaps down rather than off), may be enough.
- Yes, under the words. You need noise reduction. A gate will not touch it.
- Yes, and it changes (traffic, voices in the next room, clattering). Classic noise reduction will struggle; a speech-focused AI tool is the better choice.
Can you use both?
You can, but with a modern speech tool you rarely need to. If you do add a gate afterwards, set it to reduce the gaps by a few decibels rather than to full silence, and use slow release settings so word tails survive. Complete digital silence between sentences sounds unnatural; a little consistent room tone sounds like a real room.
Things neither tool fixes
Be realistic about what any noise tool can do:
- Echo and reverb. These are reflections of the voice itself. Noise tools don't remove them; a closer microphone and a softer room do. (See how to record clean audio at home.)
- Clipping and distortion. If the recording hit the top of the meter, the waveform is damaged. Our tool softens light clipping before it cleans, but heavy distortion can't be fully undone.
- Music you want to keep. Speech-focused tools aim to keep voices. Music under speech will be reduced, so add music after cleaning.
- Speech quieter than the noise. If you can barely make out the words yourself, any tool will be guessing.
The quick way to try it
Drop a recording into the tool, preview the original against the cleaned version at the same moment, and listen for the things a gate would have broken: word endings, breaths and the space between sentences. For video, the video noise remover replaces only the audio and leaves the picture as it is.
If you are working on a show, the podcast audio cleanup page has presets for the loudness podcast apps expect.