In this guide
If the effect has its own editor track, change that track first. For an effect mixed over speech, describe the unwanted source and review Background audio. Compare the complete sentence and surrounding scene before returning the WAV to your video; no method guarantees recovery of obscured words.
To remove a sound effect from video while keeping speech, first check whether you still have its separate track. In the original editing project, mute or lower that effect clip. In a finished video, speech, effects and music may already share one audio track. Muting that track removes them together; source isolation is a different approach that estimates the unwanted sound.
The right starting point depends on the overlap. A notification between sentences can need a short timeline edit. A door slam over a word needs a test that preserves that word. This guide focuses on reducing one unwanted event while keeping the surrounding scene, rather than extracting an effect to reuse elsewhere.
Choose the method from the recording you have
| What you have | Start here | Review before exporting |
|---|---|---|
| The effect is a separate clip in the editing project | Mute, remove or lower that clip | Whether the clip also contains wanted sound or feeds shared effects |
| An unwanted event falls in a gap between words | Edit its level locally, retaining duration when picture timing matters | Continuity of room tone, the next word and both boundaries |
| The effect overlaps wanted speech in a finished mix | Test separation of the unwanted source, then review the remainder | Missing consonants, leftover effect and changed ambience |
| Only a clean speaking voice is needed | Compare a voice-isolation workflow | Whether losing music, reactions and scene sounds is acceptable |
A stereo file is not automatically two independently editable sources. Check whether a useful microphone is already on one side with the channel and stem guide. Selecting the useful side may help a two-input recording, but it will not unmix sounds already combined on that side.
Keep the duration when editing an event between words
Listen from before the unwanted sound through the next complete phrase. If nothing wanted overlaps it, use your audio or video editor to lower that range or replace it with suitable room tone from the same recording. Review the transitions: dropping to digital silence can make a room sound disappear abruptly.
For example, suppose a notification occupies source seconds 12.20–12.65 while the speaker is silent. Removing that 0.45-second interval and closing the gap moves everything afterward earlier. Lowering the audio over the interval while leaving its duration intact preserves the later placement. These are fictional timing examples, not tested cleanup results. Choose a timeline edit that matches the picture and check the actual export.
The free audio trimmer keeps one continuous excerpt. It does not erase an interior event while retaining the rest of a video timeline. A fade also changes every sound in its selected edge; it cannot keep a word while lowering only the effect over that word. Use the cut-click guide if the edit itself introduces a new click.
For an overlapping effect, target what you want to reduce
- Keep the best original. Make a working excerpt containing the unwanted event, a complete affected phrase and surrounding ambience. Record its original start time.
- Upload a supported file. The sound-effect isolation workflow accepts MP3, WAV, MP4 and MOV up to 50 MiB (52.43 MB). A video-page URL is not an uploadable recording.
- Describe one audible source. Candidate descriptions include “the notification chime” or “the door slamming.” These are requests to test, not demonstrated successful removals. “Make everything better” does not identify the source.
- Review Background audio. When the request names the unwanted effect, this is the remaining mixture. Isolated sound is the estimated effect itself, useful for checking what the separation assigned to the target.
- Compare with the original. Listen to the same passage at similar playback levels before deciding whether to download the chosen WAV. AI downloads require an account and sufficient credits.
Requesting “spoken voice” instead changes the goal: Isolated sound becomes the voice candidate and can leave out ambience, music or reactions you wanted to retain. The prompt and output-selection guide explains this target-versus-remainder choice. Do not switch the prompt and compare a different output without noting that change.
Review the words and the scene, not just the noise
A quieter effect does not establish a better edit. Compare the beginning, overlap and decay of the event, followed by the next phrase. Check these separately:
- Speech: Are the first consonants, quiet syllables and word endings still intelligible? Has a word become thin, broken or missing?
- The unwanted event: Does its initial hit remain? Does a bell tail or other recognizable fragment continue behind later words?
- Scene continuity: Do room tone, music and wanted reactions change abruptly around the processed range?
- Other occurrences: If similar effects occur elsewhere, check them independently. One useful excerpt does not validate the entire recording.
Keep the original available for comparison. If the effect obscured a word, do not infer that an intelligible-sounding estimate is an exact recovery of the speaker’s original performance. Another microphone, a clean alternate take or a deliberate re-recording can be a better source when available. Repeated processing has no guaranteed stopping point or recovery result.
Return the remaining audio to the video
Saturalabs returns separated audio, not a finished replacement video. Import the chosen WAV into your video editor at the matching source position. If you processed an excerpt beginning at source second 30.00, its own zero refers to that excerpt’s beginning, not automatically to the beginning of the full video.
When replacing a processed interval, audition the WAV with the corresponding original interval muted so you do not add the unwanted effect back underneath it. Keep audio outside that interval as intended, and review both joins. The video audio replacement guide covers offsets, doubled sound, duration differences and sync checks.
Listen to the final rendered video around the event and near both ends. Keep a copy of the original project and recording. A good preview of a separated file does not by itself verify the timeline assembly or exported video.
Common questions
Can I remove applause while keeping a speech?
You can test applause as the unwanted target and inspect Background audio, but overlapping claps, reactions and speech can leave remnants or lose wanted detail. Compare the complete affected sentence. If the applause occupies a gap, a local timeline edit may be sufficient; if only the voice matters, compare the separate voice-isolation workflow.
Will a noise gate remove a loud effect over a word?
A level gate does not identify which sound is a word and which is an effect. A loud overlapping event can remain while quieter speech is reduced. The cleanup method guide distinguishes gating, noise reduction and source isolation.
Can I remove only the music and keep sound effects?
That is a different target. Follow the music removal workflow, target music and review the remaining mix for both speech and wanted effects. Voice-only isolation is not the same goal.
What if I want to save the effect instead?
Target the audible effect and review Isolated sound rather than Background audio. The effect extraction guide covers attack-and-tail checks, finishing a working copy and a source-timing cue sheet.