Creators, YouTube & podcasts

How to Remove Background Noise in Audacity Without Ruining Speech

Audacity's Noise Reduction effect removes steady background noise, like fan hum, computer hiss or air conditioning, by learning what the noise sounds like and subtracting it. Used gently, it makes speech clearer for listeners. Used too strongly, it leaves a watery, robotic sound that can be worse than the noise. And if the recording is only going to a transcription tool, you may not need it at all.

5 min read · Updated

What Noise Reduction can and can't fix

mydubly text to audio voice ryan clear with style sample card

Noise Reduction works on steady noise: a sound that stays roughly the same throughout. It handles well:

  • Hiss from microphones, preamps and old tapes.
  • Fan, air conditioning and computer noise.
  • Constant hum (although a dedicated hum filter is often better; see audio hum and buzz).

It does little for changing noise: traffic, other voices, a dog barking, music in the background or keyboard clicks. Trying to remove those with Noise Reduction mostly damages the voice.

Step 1: get a noise profile

mydubly text to audio voice aiden bright with style sample card
  1. Find two or more seconds where nobody speaks and only the background is audible. Many recorders capture this at the start, before anyone talks.
  2. Select just that stretch.
  3. Open Effect > Noise Removal and Repair > Noise Reduction (in older versions, Effect > Noise Reduction).
  4. Click Get Noise Profile. The window closes; Audacity has learned the noise.

A good profile is the most important step. If the selection includes a breath or a word, Audacity treats that as noise too and removes some of the voice.

Step 2: apply it gently

mydubly text to audio voice aiden energetic with style sample card
  1. Select the whole recording (Ctrl+A or Cmd+A).
  2. Reopen Noise Reduction. Start with the defaults: noise reduction of 12 dB, sensitivity 6 and frequency smoothing of 3 bands.
  3. Click Preview and listen to the speech, not the gaps. If the voice sounds hollow, lower the noise reduction value.
  4. Click OK to apply.

The Residue option plays only what's being removed. If you can hear words in the residue, the settings are too strong.

Signs you've overdone it

mydubly text to audio voice ryan angry with style sample card
  • Speech sounds underwater, metallic or "chirpy" (often called musical noise).
  • Consonants like s, t and f lose their edges.
  • The background pumps in and out between words.

Undo (Ctrl+Z) and try a lower reduction. Two gentle passes with a fresh profile can sound better than one strong pass, but each pass costs some voice quality. Learning the basics of EQ for voice recordings helps too: a gentle high-pass filter removes rumble without any noise profile.

Example: a lecture recorded near a projector

A student records a lecture on her phone, with projector fan noise throughout. For her own listening, she takes a noise profile from the 3 seconds before the lecturer starts, applies 9 dB of reduction, and the hiss drops without the voice going thin. For the transcript, she uploads the original recording, since the fan noise is steady and doesn't affect it.

Do you need noise reduction before transcription?

mydubly text to audio voice ryan bedtime with style sample card

Often not. Modern speech recognition models like Whisper are trained on huge amounts of real-world audio, much of it noisy, and handle steady background noise well. Heavy noise reduction can make things worse for these models, because artifacts confuse them more than hiss does. A useful rule:

  • Steady, moderate noise: upload the original.
  • Loud noise that drowns words: try gentle reduction, then compare transcripts of a short excerpt from both versions.
  • Changing noise (crowds, traffic): noise reduction won't help much; the transcript will need more proofreading.

Speech recognition and background noise explains how models cope, and transcribing quiet audio covers recordings where the voice is too soft.

Cleaning audio for a dub or a mydubly transcript

mydubly text to audio voice ryan horror with style sample card

For a transcript, upload the original or cleaned file to audio to text; export from Audacity as WAV or MP3 with File > Export Audio. For a dubbed video, the dub copies each speaker's voice from the original recording, so a cleaner voice gives a cleaner copy. Work on the audio in Audacity, then put it back on the video in your editor before uploading. The dub keeps background music, but steady hiss is part of the voice signal and is worth reducing first.

Other Audacity tools worth knowing

Noise Reduction is one of several effects that help speech recordings:

  • High-Pass Filter (Effect > EQ and Filters) removes low rumble from traffic, handling noise and air conditioning. A cutoff around 80 to 100 Hz rarely touches a voice.
  • Normalize brings quiet recordings up to a consistent peak level without distorting them.
  • Compressor evens out a speaker who drifts toward and away from the microphone.
  • Click Removal helps with old recordings that crackle.

Apply them in a sensible order: high-pass first, then noise reduction, then compression and normalisation. Save a copy of the original before you start, so you can always go back. For lecture recordings and interviews that go straight to a transcript, the lecture transcription page covers the student workflow, where a high-pass filter is often the only cleanup worth doing.

Prevention beats repair

Noise reduction is a rescue tool. Turning off the fan, recording closer to the microphone and choosing a quieter room give you better audio than any effect. Room acoustics for voice recording covers cheap fixes like recording in a room with soft furnishings, and audio recording settings for speech covers the recorder side.

Frequently asked questions

Where is Noise Reduction in Audacity?

Effect > Noise Removal and Repair > Noise Reduction in current versions; Effect > Noise Reduction in older ones.

What are good Audacity noise reduction settings for voice?

Start with the defaults (12 dB, sensitivity 6, frequency smoothing 3) and lower the reduction if the voice sounds hollow.

Why does my voice sound robotic after noise reduction?

The reduction is too strong or the noise profile included speech. Undo, take a cleaner profile and use a lower setting.

Can Audacity remove background voices or music?

Not well. Noise Reduction is designed for steady noise like hiss and fans; changing sounds stay or damage the voice when removed.

Should I remove noise before transcribing?

Usually not for steady, moderate noise. Modern models handle it, and artifacts from heavy reduction can hurt accuracy.

How do I export from Audacity for transcription?

File > Export Audio, as WAV or MP3. Either uploads directly to mydubly.