What you’ll be able to do

You know that recording you made in your kitchen with the fridge humming in the background? Or the interview where the room echoes so much it sounds like a cave? You can fix that. Not perfectly, not always, but a lot.

The tools I’m covering here strip out background noise, cut echo, and generally make a recording sound like it was made somewhere quieter and better treated than it actually was. No audio degree needed, no expensive plugins. Just a browser and a file.

It won’t turn a phone recording made in a stairwell into a broadcast studio take. But it’ll get you a lot closer, and for most podcasts, voice notes, and interviews, that’s exactly enough.

Before you start

A few things worth knowing before you upload anything.

First, consent. If you’re recording someone else, cleaning up the audio doesn’t change the rules about whether you had permission to record them in the first place. That still applies.

Second, enhancement changes how a voice sounds. Sometimes barely noticeably, sometimes quite a lot if you push it hard. If the recording matters as evidence, or it’s an interview where accuracy of tone matters, say so, and maybe keep a copy of the untouched original handy so you can point back to it. Courts and journalists both care about this kind of thing for good reason.

Third: the best fix happens before you even hit record. A closer mic and a softer room (think cushions, curtains, not an empty tiled kitchen) will do more for your sound than any AI tool afterwards. Enhancement is a patch, not a replacement for decent recording conditions. Keep that in mind and you’ll get better results, faster, with less fiddling.

And always, always keep your original file. Never overwrite it. If the enhanced version sounds worse or weird, you want to be able to start again.

Get set up

For most people, start with Adobe Podcast Enhance. It’s free, it runs in your browser, and you don’t need an account to try it. You upload a file, it processes it, you download the result. That’s genuinely it. It’s built specifically to strip noise and echo from speech, which is exactly the job most people need doing.

If you’re already editing inside Descript, its Studio Sound feature does a very similar job without you needing to leave the editor. Handy if you’re already cutting a podcast there and don’t want to bounce files between apps.

For more control, Audacity (also free) has a noise reduction tool built in, but it works differently: you need to give it a sample of just the noise, a few seconds of “silence” from your recording where nobody’s talking, so it learns what to remove. A bit more manual, a bit more fiddly, but it’s free and it’s been around for years.

If you’re working with video rather than pure audio, CapCut and DaVinci Resolve both have denoise tools built into their editors, so you don’t need to hop out to a separate audio tool if you’re already cutting there.

ElevenLabs has an audio isolation tool worth knowing about too, particularly if you’re trying to separate a voice from music or other complex background sound rather than just hiss and hum.

And if you’re doing this professionally, or the recording is genuinely important (a client project, a big interview), iZotope RX is the industry standard. It costs money and has a learning curve, but it does things the free tools can’t.

For a first go, Adobe Podcast Enhance is where I’d point you. Free, fast, no learning curve.

Try it yourself

Here’s how to actually do it, step by step, with a few different approaches depending on what you’re working with.

Using Adobe Podcast Enhance directly. Go to the tool, upload your audio file, let it process. There’s no prompt here, it’s not that kind of tool, but here’s what you should be checking as it works:

Before uploading: confirm the file is a copy, not your only version.
After processing: listen to the first 10 seconds and the last 10 seconds separately.
Compare: does the voice sound natural, or slightly "underwater"?

Using an AI assistant to help you plan the clean-up. If you’re not sure which tool suits your recording, or how to describe the problem, this is where something like ChatGPT, Claude, or Gemini is genuinely useful, not for the audio processing itself, but for figuring out your approach.

I have a podcast recording made on a laptop mic in a small room with 
some echo and a fan running in the background. I want it to sound 
cleaner without sounding robotic. What order should I do things in, 
and which free tools should I try first?

Using an AI assistant to write your Audacity noise sample instructions. If you’re going the Audacity route and haven’t used it before, ask for plain steps:

Explain, in simple steps a complete beginner could follow, how to use 
Audacity's noise reduction tool: how to select a noise-only sample, 
how much reduction to apply, and how to avoid making the voice sound 
distorted.

None of these AI assistants can process your audio file directly inside the chat the way Adobe Podcast Enhance or Descript can. Think of them as a second opinion or a set of instructions, not the tool doing the actual cleaning.

Check the result

Don’t just glance at the waveform and assume it’s fine. Listen. Properly.

Play the enhanced file next to the original, back to back, ideally on headphones rather than laptop speakers. Laptop speakers hide a lot of sins.

Listen specifically for these things:

The “underwater” effect: a slightly muffled, swimming quality to the voice, especially on esses and softer consonants. This happens when noise reduction is pushed too hard and it starts eating into the voice itself, not just the noise around it.

Robotic or metallic artefacts: a thin, slightly synthetic edge to the sound, particularly noticeable on vowels. Another sign of over-processing.

Pumping: where the background noise seems to swell and fade unnaturally between words, like it’s breathing. This is common with automatic noise gates and less so with AI enhancement, but still worth a listen.

Play it to someone else if you can. Fresh ears catch artefacts you’ve stopped hearing after listening to the same ten seconds forty times.

A simple test that works well: play the enhanced version to someone who’s never heard the original and ask them where they think it was recorded. If they guess “studio” or “quiet room,” you’ve done well. If they say “phone call” or hesitate, it needs another pass or a different tool.

If it doesn’t work

Sometimes enhancement makes things worse rather than better. Usually it’s because the noise reduction was pushed too hard, or the original recording was too far gone to begin with. AI can’t invent detail that was never captured; if the voice is buried under noise on the original file, no tool fully recovers it.

If the result sounds artificial, try a lighter setting rather than the maximum. Most tools including Adobe Podcast Enhance don’t offer sliders, in which case try a different tool that does, like Audacity, where you control the strength yourself.

If nothing sounds right, go back to the source. Can you re-record in a better spot? A wardrobe full of clothes, oddly enough, makes a brilliant vocal booth if you’re stuck. Blankets over a table works too. It sounds daft, but it works, and it’ll save you more editing time than any plugin.

On privacy: don’t upload recordings containing other people’s personal information, financial details, medical information, or anything confidential to a general-purpose tool unless you’ve checked its terms and you trust it with that data. Recordings of children, sensitive interviews, or anything under an NDA need extra care, not a default upload to whatever tool is fastest. When in doubt, don’t upload it, ask first, or use a tool your organisation has already approved for that kind of content.

Keep exploring

If you’re working with recordings regularly, a few other guides on this site build on what’s here: “How to get AI to transcribe an audio recording” is a natural next step once your audio’s clean. If you need to add narration rather than clean up existing speech, see “How to get AI to make a voice-over.” And if your project involves video rather than pure audio, “How to get AI to edit a video” covers the wider process.

Sources and review notes

Review date: 15 September 2026 — every source above was opened and checked against the text on that date.