In this guide
Start by isolating the voice you want to keep. Compare the result with the original at a similar volume, and listen for missing words before you decide that a quieter track is a better track.
A good interview can happen in a noisy room. A fan, passing traffic, or music from the next table can make the recording harder to follow. The goal of cleanup is to make the words easier to understand while keeping the speaker natural.
Different noises need different approaches. Before processing a full video, listen to a short section that includes both speech and a pause. That gives you a useful test for what the cleanup removes and what it damages. The audio isolation and noise reduction comparison helps you choose whether to test a profile-based tool, source isolation, a gate, or a small local edit.
Identify what is competing with the voice
First decide whether you are hearing a steady background, another distinct sound, or a recording problem. This helps you choose a method instead of applying the same treatment to every clip.
| What you hear | A useful starting point | What to check |
|---|---|---|
| A steady fan or hiss | Noise reduction using a noise-only sample | Thin or metallic speech after processing |
| Music, traffic, or other overlapping sounds | Isolate the spoken voice | Words disappearing where sounds overlap |
| Clicks between sentences | Small manual edits or fades in an editor | Abrupt jumps in room tone |
| Distorted, overloaded speech | Look for a cleaner original or alternate recording | Distortion already present before cleanup |
Audacity's Noise Reduction documentation distinguishes steady noise from irregular sounds and explains the tradeoff between suppression and damage to the wanted audio. Source isolation is a different approach: you describe the sound to extract rather than teaching the tool a noise profile.
Isolate dialogue in Saturalabs
- Start with the original file. Open the background noise removal tool and upload your recording. The upload interface supports MP3, WAV, MP4, and MOV files up to 50 MiB (52.43 MB). If the video is larger, make a short test clip in your video editor first.
- Describe the sound to keep. Try “spoken voice” or “the person speaking.” These are starting points, not guaranteed settings. Avoid mixing a target with a long list of instructions.
- Select “Isolate sound.” Once processing completes, compare the original, isolated sound, and background audio. The isolated track should contain your target; the background track contains what remains.
- Listen before downloading. Check consonants, quiet words, pauses, and places where a loud noise overlaps the speaker. The download dialog shows the credit cost before you download.
If you instead isolate an unwanted sound, such as a dog barking, the track to inspect is Background audio. That is the version with the isolated target removed. Always audition it: some wanted speech may have been removed too.
Compare intelligibility, not just silence
A track can sound impressively quiet between sentences and still lose the beginnings or endings of words. Alternate between the original and processed versions at a similar listening volume. A louder version can seem clearer even when it is not.
- Listen to a complete sentence without looking at a transcript. Can you still understand each word?
- Check breaths and soft consonants. Do they disappear or sound unnaturally sharp?
- Listen where the noise becomes loudest. Does the speaker become watery or hollow?
- Try the result on both headphones and the speakers your audience is likely to use.
If the voice is damaged, try a simpler description on the original file. Keep your first result so you can compare attempts. Repeatedly processing an already processed track can make it harder to tell which step caused a problem.
Bring the audio back into your video edit
Download the version you chose, then import it into your video editor. Place it at the same starting point as the source audio and mute the original track so the two versions do not play together.
Check lip sync at the beginning and end. If you tested a trimmed excerpt, align it with that same excerpt in the timeline. Add any fades or background ambience in the editor and listen to the final video export, not only the audio preview.
Follow the guide to replacing video audio for full-length and trimmed WAV placement, doubled dialogue, sync checks, and an optional FFmpeg workflow.
Know when to stop
Some room sound is preferable to a damaged voice. If your dialogue is understandable and the remaining noise is unobtrusive in the finished video, further processing may not improve it. A clipped or heavily obscured voice may need an alternate take. Even dedicated repair tools have limits: Audacity describes Clip Fix as most likely to help lightly clipped audio.
Common questions
Can I remove music while keeping dialogue?
Try isolating “spoken voice” and inspect the isolated track. Use the music removal workflow when music is the main distraction. Similar voices or instruments overlapping the dialogue can make clean separation difficult.
Will this fix every noisy recording?
No. Source quality and overlap matter. Test the hardest section before committing to a workflow, and keep the original as a reference.
What should I change for the next recording?
Move closer to the microphone, find a quieter position, and record a short playback test before starting. Better separation between the voice and its surroundings at recording time gives cleanup less work to do.
