Audacity's Noise Reduction effect asks for a noise profile: a selection containing the unwanted sound without the voice you want to keep. If you cannot find one, resist selecting a word just to fill the requirement. The effect can mistake wanted speech for noise and remove part of it.
Listen for a noise-only pause. If there isn't one, is the background steady beneath continuous speech, or does the voice itself sound damaged? That distinction changes what you can try.
Is there a genuine noise-only gap?
Listen at the start and end of the recording and between sentences. A small pause may contain a representative fan or hiss with no speech, music or handling sound. Zoom in on the waveform, select that gap, and play the selection on its own. Do not rely only on a visually quiet waveform: a breath or soft syllable can be easy to miss.
Audacity's Noise Reduction manual describes the two-pass process: get a profile from noise alone, then select the audio to treat and preview the effect. Save an unchanged copy, use the preview and listen for words becoming dull or watery. Menus and controls can vary by version; consult the instructions for the version you use.
What if speech covers every part of the recording?
If the background is steady but there is no noise-only selection, the ordinary profile workflow has no clean example to learn from. Do not use the first clean speech as if it were noise. You can look for an unused pause from the same recording setup, provided it contains the same noise and no wanted sound. If you cannot find one, consider another tool or ask for the original recording.
For a supported short file, the browser noise remover offers a preview without asking you to select a profile. The input must fit both the 25 MiB and five-minute limits; the preview covers up to 30 seconds. There is no strength slider. Listen before keeping the result: a different method will not always sound better.
What if the buzz is present only when someone speaks?
That can indicate a different problem from a fan running continuously. Listen for clipping, compression, a gate opening, connection artifacts or another process already applied to the recording. You cannot determine the cause merely from the symptom. If possible, request an untreated source and check whether the sound is present there.
One Audacity forum discussion describes buzz during speech with quiet gaps in between. It is a useful example of the question, not proof that every similar file has the same cause or can be repaired automatically.
How do you decide whether the result helped?
- Keep the original and prepare a copy of a passage containing the difficult speech.
- Listen to the same sentence before and after at the same playback volume.
- Check quiet consonants, word endings and breathing before judging the pause.
- Keep the processed version only if the words remain natural and the distraction is less bothersome.
If the recording starts with a clean introduction but the problem occurs later, prepare a separate excerpt in an editor first. This site previews the beginning of the selected file; it does not provide a later-point selector before processing. Our controlled examples show how original and processed clips can be compared, but they are not a test of your file.
Common questions about noise profiles
Can I profile a part that contains voice?
It risks teaching the effect to treat that voice as unwanted noise. Choose a noise-only passage for the documented profile workflow.
Can a noise remover restore clipped words?
Not reliably. Preserve the original and seek a different source or a repair tool designed for that problem.
Should I process the same clip repeatedly?
Listen after the first pass. Stop if additional processing makes the voice less natural or less intelligible.