How Do You Rescue Bad Audio Without Recording It Again?
The best tool for a ruined recording costs nothing and takes thirty seconds. Here is what it fixes, what it cannot, and what paying actually buys you.
By Merxtio Staff
The interview is done, the guest has gone, and the recording has a refrigerator in it. Or a laptop fan, or a room with too many hard surfaces, or a microphone that was pointing the wrong way for forty minutes.
You can almost certainly fix it, and the tool that does most of the work is free. This page covers what free handles, the three different jobs that all get called "AI audio", and what paying actually buys. About six minutes.
Fix it for nothing first
Adobe Podcast's Enhance Speech does one thing and does it better than tools costing real money: it takes recorded speech and strips the room out of it. Background hum, echo, air conditioning, the general muddiness of a laptop microphone.
Upload the file, wait, download it. No account needed for a single upload. The free tier handles roughly an hour of processing a day, with a limit of 30 minutes and 500MB per file — which covers most interviews and every voiceover.
Run this before you consider buying anything. If it fixes the recording, you are finished, and the rest of this page is a reference for next time.
What it will not fix: two people talking over each other, a section where the microphone was off, or a recording so clipped that the words themselves are gone. Enhancement recovers what was captured badly. It cannot recover what was never captured.
The three jobs
| Feature | The job | What it costs to start |
|---|---|---|
| Cleanup | Removing noise, echo and room from speech that was recorded badly. | Free. Paid tiers add batch processing and longer files, not better cleanup. |
| Editing | Cutting, removing filler words, and assembling the finished thing. | From $12 a month, and this is where the hours actually go. |
| Synthesis | Generating speech that nobody recorded, or cloning a voice. | Free tier exists; realistic use starts around $22 a month. |
The categories fail differently, which is the useful part. Cleanup either works or the audio was unrecoverable. Editing always works and always takes time. Synthesis works technically and raises questions about disclosure that the other two do not.
What paying buys
| Item | USD per month, annual billing where offered |
|---|---|
| Cleanup, free tier | Free |
| Adobe Podcast Premium | $9.99 |
| Descript Creator, editing | $12 |
| ElevenLabs Creator, synthesis | $18.33 |
Notice what the paid cleanup tier actually adds. Adobe Podcast Premium at $9.99 a month buys batch uploads, video support and files up to two hours — throughput, not quality. If you clean one file a week, the free tier is not a limited version of the paid one. It is the same tool with a queue.
Auphonic prices differently and is worth knowing about if volume is your problem: free covers two hours a month, then $13 for nine hours, $27 for twenty-one, and up from there. Paying by processing hour rather than by seat is unusual, and it suits an irregular podcast better than a monthly subscription does.
When synthesis is the right tool
Generating a voice is a different decision, with a different set of questions attached.
That last point is worth stating plainly. Cloning your own voice to patch a mistake is an editing convenience. Cloning someone else's without asking is not, whatever the tool permits, and it is the fastest way to turn a production shortcut into a problem.
The order to work in
Enhance the raw file before you edit anything
You’ll have: A clean master to edit from. · about 1 minute per file
Run the original through a free enhancer first, then edit the enhanced version. Doing it the other way round means enhancing each cut separately, and the processing is not always identical between them.
Keep the original. Enhancement is destructive in the sense that it makes choices for you, and occasionally it makes one you dislike on a particular voice.
Cut the content before polishing the sound
You’ll have: A rough edit at final length. · about Half the time you expect
Remove the tangents, the false starts and the three minutes about parking. This is where the finished thing is actually made, and no amount of audio processing substitutes for it.
Doing it in a transcript rather than a waveform is the largest single time saving available in audio work, because reading is faster than scrubbing.
Strip filler words automatically, then listen back
You’ll have: Speech that sounds considered rather than clipped. · about 10 minutes
Automatic filler removal is genuinely good and slightly overzealous. Removing every single one produces a delivery that sounds oddly relentless, because natural speech has hesitation in it.
Listen to a minute of the result before accepting all of them. Keeping some back is usually the better call.
Only then consider paying for anything
You’ll have: A subscription bought against a named limit rather than a hope. · about 15 minutes
By now you know which step was the bottleneck. If cleanup was fine and the editing took the day, the money belongs in an editor. If you were queueing files behind a daily limit, it belongs in throughput. If you needed narration that never existed, it belongs in synthesis.
If the recording is destined for video, the same tool covers both and the video guide sets out where the categories overlap. And the four-question test in the tool selection guide matters unusually much here, because audio tools are bought after one disaster and used twice.
Questions people ask
- How do I fix a bad audio recording?
- Run it through a free speech enhancer before doing anything else. Adobe Podcast Enhance Speech removes room noise, echo and hum from recorded speech in about thirty seconds and needs no account for a single file. That solves most bad recordings outright.
- Is Adobe Podcast Enhance free?
- Yes. The core Enhance Speech tool is free, with roughly an hour of processing a day and a limit of 30 minutes and 500MB per file. Premium is $9.99 a month or $99.99 a year, and what it adds is batch uploads, video support and longer files — throughput rather than better cleanup.
- Can AI fix audio recorded on a phone?
- Usually, and this is the case where it looks like magic. A phone recording of speech in an ordinary room comes back close to broadcast quality. What it cannot recover is words that were never captured — heavy clipping, a dead microphone, or two people talking over each other.
- How do I remove um and uh from a podcast?
- Transcript-based editors do it as a single action. Descript from $12 a month detects filler words and removes them in bulk. Listen back before accepting every one, though — stripping all hesitation makes delivery sound unnaturally relentless.
- What is the best AI voice generator?
- ElevenLabs is the quality reference, with a free tier and a realistic entry point at Creator, $22 a month or $18.33 billed annually. Judge it on your own script before paying: whether a synthetic voice passes depends far more on your audience than on the benchmark.
- How do I remove background noise without expensive software?
- You do not need any. The free tier of a dedicated speech enhancer outperforms most paid noise reduction on recorded speech, because it is a purpose-built model rather than a general filter. Pay only when you need to process many files or long ones.