By the end of this guide you will have a cleaner copy of a voice recording, with background noise such as hiss, hum, traffic or echo reduced, and you will know how to check that the speech still sounds natural. The job is done by an AI noise-removal tool: you upload the file, the tool separates the voice from the noise, and you compare the result with the original before saving it.
If you want to remove background noise from audio with AI, the main choice is where the cleanup should live: in a dedicated cleaner, inside an editor you already use, or as one step in a podcast or video workflow. This guide walks through the steps and compares six tools, from our Audio and voice category and neighbouring ones.
Key takeaways
- Keep the original, upload a copy to a noise-removal tool, set the strength, compare before and after, then export a clean file.
- The main decision is the kind of tool: a one-purpose cleaner, an editor with built-in cleanup such as Descript or Kapwing, or a podcast-focused tool such as Resound.
- Free allowances are small and have caps on file length and upload size, so check them against your recording first.
- AI cleanup cannot fix everything. Overlapping speakers, music in the voice range and severe clipping are listed as outside what at least one tool can repair.
What you need before you start
- The recording. A voice file such as MP3, WAV or M4A, or a video file with speech in it.
- A backup. Keep the untouched original so you can start again if the result sounds wrong.
- Headphones. You cannot judge a cleanup on laptop speakers.
- Time. Under an hour for a typical recording, including a second try.
- Difficulty. Easy. Noise Reducer and LALAL.AI list an adjustable strength or noise-canceling level, and Resound describes mixing and mastering in one click.
- An account, sometimes. Noise Reducer lets you preview before you create an account. Check the sign-up terms of the others.
Which kind of AI tool fits this job
Pick by where the recording came from and what you plan to do with it next.
A dedicated cleaner. Noise Reducer works in the browser on audio and video files, targets hiss, hum, wind, traffic, echo and chatter, and lets you compare the original with the cleaned version before downloading. LALAL.AI is mainly a stem splitter, but its Voice Cleaner removes background noise with an adjustable Noise Canceling Level, and it also has de-echo and de-reverb options. It runs on the web, desktop and mobile.
An editor with cleanup built in. Descript edits audio and video by editing the transcript, and lists Studio Sound for audio enhancement. Kapwing is a browser-based video editor with Clean Audio and Auto Level Audio, useful when the noisy recording is part of a video project.
An audio platform with a voice isolator. ElevenLabs lists a Voice Isolator to clean up dialogue as part of a wider audio platform that also does text to speech, dubbing and transcription.
A podcast workflow tool. Resound finds ums, ahs and long silences, and its Enhance feature mixes and masters the audio with AI, removing background noise and evening out levels. It does not record or transcribe audio.
Step 1: Back up the original and listen for the problem
Knowing what kind of noise you have tells you which tool and setting to use.
- Copy the file to a safe folder and work on the copy.
- Listen on headphones and name the noise: constant hiss, low hum, wind, traffic, echo or other voices.
- Note whether the noise is steady or comes and goes.
- Mark any section where someone talks over someone else, or where music is mixed in, because those are the hard cases.
Check it worked by being able to describe the problem in a sentence, such as "steady hiss and a little room echo".
Step 2: Pick a tool and check its limits
You end this step with a tool whose allowance fits your file.
- Compare file length and size against the free allowance. Noise Reducer's free processing covers files up to 20 minutes long, within a 120-minute allowance per rolling 24 hours, and its audio upload limit is 500 MB.
- LALAL.AI's Starter plan includes 10 minutes and a 200MB upload limit per file. Resound's free tier covers 20 minutes of processed audio.
- For video, check that the tool works on video directly. Noise Reducer returns video with the picture untouched, and Kapwing is a video editor.
- Use the comparison table below for pricing and stated limits.
Check it worked by confirming your recording fits inside the free allowance or the plan you chose.
Step 3: Upload the file and run the cleanup
The goal is a first cleaned version with the strength set to a moderate level.
- Upload the copy, or record it directly if the tool has a recorder. Noise Reducer has a built-in browser recorder.
- Start with a moderate strength rather than the maximum. Noise Reducer offers adjustable strength, and LALAL.AI's Voice Cleaner has an adjustable Noise Canceling Level.
- If the recording has echo, turn on de-echo or de-reverb options where they exist, as in LALAL.AI. Noise Reducer notes that heavy reverb is reduced rather than removed.
- Run the process and wait for the preview.
Check it worked by hearing a preview that is quieter in the gaps between words.
Step 4: Compare before and after
Listen critically, because overprocessing makes voices sound thin or watery.
- Play the original and the cleaned version back to back on the same section.
- Listen to the quiet moments for leftover hiss, and to the voice for a metallic or underwater sound.
- If the speech sounds processed, lower the strength and run it again.
- If the noise is still there, raise the strength a little, or try a different tool, since each uses its own model. LALAL.AI lists a choice between neural networks on the same track.
Check it worked by being unable to tell which clip is which when you play only the speech.
Step 5: Level the audio and fix remaining issues
Cleaning often leaves the volume uneven.
- Even out loud and quiet parts. Resound's Enhance evens out levels toward streaming loudness standards, Kapwing lists Auto Level Audio, and Noise Reducer's speech enhancer evens out volume and rebalances frequencies.
- Fix remaining clicks, pops or clipping. Noise Reducer lists repair of clipping, pops and compression artifacts, though it says severe clipping is outside what it can fix.
- In an editor such as Descript, remove filler words and long gaps while you are there.
- Listen to the full file once from start to finish.
Check it worked by playing the full recording at normal volume without needing to adjust the volume control.
Step 6: Export and keep both versions
Finish by saving the clean file in a format that suits where it will be used.
- Export in the format you need. LALAL.AI exports MP3, WAV, FLAC, OGG or AAC, and Resound exports MP3, WAV, AAF and MP4.
- Save the cleaned file next to the original, with clear names.
- Download promptly. Resound deletes exported projects after 1 to 60 days depending on the plan.
- Use the cleaned file in your video, podcast or call recording.
Check it worked by opening the exported file in your editing app and playing it against the picture or the other tracks.
Tools for this job compared
| Tool | Good fit for | Pricing | Stated limits |
|---|---|---|---|
| Descript | Cleaning audio while editing by transcript (web) | Freemium, paid from $16/mo (checked 24 Oct 2026) | Free plan: 60 media minutes a month; unused minutes and credits do not roll over |
| Noise Reducer | Quick browser cleanup of audio and video (web) | Freemium | Free: files up to 20 minutes within 120 minutes per 24 hours; 500 MB audio upload limit; no fix for overlapping speakers |
| LALAL.AI | Voice cleaning plus stem splitting (web, desktop, mobile, API) | Freemium, paid from $7.5/mo (checked 24 Oct 2026) | Starter: 10 minutes and a 200MB upload limit per file; VST plugin and API access are Pro only |
| ElevenLabs | Voice isolation inside a wider audio platform (web, API) | Freemium, paid from $6/mo (checked 24 Oct 2026) | Free plan: 10,000 credits per month, no rollover; creative products share one credit pool |
| Resound | Podcast cleanup, filler removal and mastering (web) | Freemium, paid from $15/mo (checked 24 Oct 2026) | Free tier: 20 minutes of processed audio; no recording or transcription; exports deleted after 1 to 60 days |
| Kapwing | Cleaning audio inside video projects (web) | Freemium, paid from $16/mo (checked 24 Oct 2026) | Free exports are watermarked, up to 4 minutes and 720p; Free plan has 10 credits |
Prices and plan limits change, so confirm the current terms on each vendor's site before you subscribe.
Common mistakes to avoid
- Overwriting the original. Once a noisy file is replaced you cannot try again. Fix it by always working on a copy.
- Pushing the strength to maximum. Aggressive settings can leave voices thin or artificial. Fix it by starting moderate and raising it only as needed.
- Expecting a fix for everything. Noise Reducer lists overlapping speakers, background music in the voice range and severe clipping as outside what it can fix. Fix it by re-recording when the source is too damaged.
- Ignoring file limits. A long recording can exceed a free cap partway through. Fix it by checking length and size before you upload, or by splitting the file.
- Judging on laptop speakers. Small speakers hide artefacts. Fix it by checking on headphones before you export.
How we put this together
This is an evidence-based guide built from each tool's listing in our directory, which is compiled from the vendor's own site and pricing page. It is not a review based on using the tools. We don't rank tools, and nobody paid to be included. Some outbound links may be affiliate links; see our disclosure.
Frequently asked questions
Can I remove background noise from audio for free?
Yes, within limits. Noise Reducer's free use is capped at 120 minutes of processing in any rolling 24-hour window, LALAL.AI has an always-free Starter plan with 10 minutes, and Resound's free tier covers 20 minutes of processed audio. Full downloads on some tools need a balance or paid plan, so check before you start.
Can AI remove echo and reverb?
It can reduce them. LALAL.AI lists de-echo and de-reverb options, and Noise Reducer lists echo removal but says heavy reverb is reduced rather than removed. Expect an improvement, not a perfect dry recording.
Can I clean up audio on my phone?
LALAL.AI lists iOS and Android apps. Noise Reducer is listed as a web tool, so test it in your phone's browser before you rely on it. Check each vendor's site for current mobile features.
Can AI fix a recording with two people talking at once?
Not reliably. Noise Reducer says overlapping speakers and background music in the voice range are outside what it can fix. For those recordings, re-recording is usually the safer route.
How long does it take?
The listings give no standard processing time, so run a short clip first. Expect most of the time to go on listening and comparing.
Compare more options in our Audio and voice category, or see how to transcribe a podcast with AI for the next step in a podcast workflow.
