How to transcribe noisy audio (and what to expect)
Short answer
Noisy audio can be transcribed, but expect gaps and wrong words in the noisy stretches; no tool fully fixes it. Steady hum can be reduced first with Audacity's free Noise Reduction. Afterward, use each line's timestamp to recheck names, numbers and quotes against the audio.
A recording from a busy café, a class with the air conditioning roaring, or a meeting where everyone talks at once can still be transcribed. It just won't come out as clean as a quiet one, and no tool, free or paid, fully fixes that.
This guide explains what each kind of noise does to a transcript, what you can try before uploading, and how to check the passages most likely to be wrong without listening to the whole thing again.
What each problem does to the text
| Problem | What you'll usually see |
|---|---|
| Steady noise (fan, AC, traffic hum) | Mostly fine if voices are clearly louder; quiet words and endings get dropped |
| Echo (big, empty, hard rooms) | Words blur together; more wrong words, especially from people far from the mic |
| Background music or TV | Song lyrics or TV dialogue can slip into the transcript as if someone said them |
| Crosstalk (people talking over each other) | Short phrases go missing and lines can be attributed to the wrong speaker |
| Distant or soft-spoken speakers | Gaps or guesses where that person talks |
Before you upload: free cleanup
Noise reduction works best on steady, constant noise like hum, hiss or a fan. It does little for irregular noise like traffic, clattering dishes or other conversations, and overdoing it makes voices sound watery or robotic, which can hurt a transcript more than the noise did.
Audacity is free for Windows, Mac and Linux and has a well-documented noise reduction effect:
- Open the file and select a few seconds where there's only background noise, no one talking.
- Go to Effect → Noise Removal and Repair → Noise Reduction (older versions list it directly under Effect) and click Get Noise Profile.
- Select the whole recording, open Noise Reduction again and start with Audacity's suggested values (Noise reduction 12 dB, Sensitivity 6, Frequency smoothing 3). Click Preview, then OK.
- Listen to a minute of the result. If voices sound metallic, undo and use a lower setting.
- Export as mp3, wav or flac and upload that file.
A note on phone "enhance" buttons
On iPhone, Voice Memos has an Enhance Recording option, in the recording's options, that reduces background noise and echo. You can switch it off at any time, and it can make listening easier, but don't count on it to rescue a transcript. If a recording matters, fixing it at the source (moving closer, choosing a quieter room) is worth far more than any button afterward.
What to expect from the transcript
With Dolcenota, upload the file as usual from your browser; there's nothing special to switch on. Speakers are separated automatically and every line gets a timestamp, like [00:12:31] Speaker 2: …. Expect the quiet, clear stretches to read well and the noisy ones to have more gaps and wrong words. The summary and the answers to your questions come from the transcript, so they're only as reliable as the passages they rest on.
How to check the risky passages
- Skim for lines that don't make sense, sudden topic jumps or a speaker who seems to answer themselves.
- Use the timestamp on that line to jump to the exact moment in the audio and listen to just that part.
- Double-check names, numbers, dates and anything you'll quote; those are the costliest mistakes.
- If a speaker got split into two labels, merge them; you can undo it.
- Export to Word and correct the text there. Every export notes that it's AI-generated and may contain errors.
Frequently asked questions
Can AI remove background noise completely?
No. Noise reduction can make steady noise quieter, but it can't recover words the microphone never captured clearly, and heavy settings damage the voices too.
Should I clean the audio before uploading?
Only if there's a steady hum or hiss and the cleaned version clearly sounds better to you. If you're unsure, upload the original.
Why did song lyrics show up in my transcript?
Speech recognition can't always tell a voice on a speaker from a person in the room. Skip or delete those lines when you edit the export.
Does Dolcenota clean the audio for me?
No. It transcribes what's in the recording. For the best result, reduce the noise when you record.
Will a noisy recording use more minutes?
No. Minutes are based on the recording's length, not on how hard it is to transcribe.
Try it with your recording
Upload your audio and see the transcript, with a timestamp on every line to check the hard parts. Free to start.
Start free