How to translate a video while keeping the background music

Why most translation tools flatten your audio

Many video translation tools treat the audio as one inseparable track. When they generate dubbed speech, they either replace the whole soundtrack or talk over it, so the original music, sound effects, and ambient room tone get muffled or lost entirely.

For a talking-head clip that may be acceptable. For a music video, a vlog with a soundtrack, an ad, or a tutorial with ambient demonstration sound, losing the background audio changes how the video feels — and viewers notice immediately.

The fix: separate speech from background before you translate

The reliable way to keep the music is to split the audio into two stems first: the spoken voice, and everything else (music, effects, ambience). You translate and re-voice only the speech stem, then mix the new dubbed speech back over the untouched background stem.

Because the background stem is never regenerated, the music and sound effects stay as they were in the original — only the spoken words change language. The one caveat is the separation itself, which is an estimate: where speech and music overlap closely, some of one can end up in the other (more on that below).

How WaveShift preserves background audio, step by step

WaveShift runs this separate-translate-remix pipeline automatically. You do not configure stems or touch an audio editor:

• Add your video by uploading a file or pasting a YouTube, Bilibili, Twitch, or X (Twitter) link.
• WaveShift separates the speech from the background music and sound effects.
• It translates the speech and clones each speaker's voice so the dubbed audio keeps the original speaker identity where possible.
• It mixes the translated speech back over the original background audio, so music and effects stay intact.
• Playback streams as it renders — you hear the first dubbed segment in minutes while the rest continues.
• If one line sounds off, you can edit that single subtitle line and regenerate only that line, without redoing the whole video.

When keeping the background matters most

Background preservation is the difference between a usable localized video and one you have to re-edit. It matters most for:

• Music videos and performances, where the track is the point.
• Vlogs and lifestyle content with a continuous soundtrack.
• Ads and trailers, where sound design carries the mood.
• Tutorials and product demos with ambient or on-screen sound.

Tips for the cleanest result

A few habits make the separated-and-remixed output sound natural:

• Start from a source where the speech is reasonably clear over the music — cleaner input separates better.
• Review one representative segment first, including a line with music underneath it, before processing a long video.
• Use single-line re-dubbing to fix any line that lands awkwardly instead of regenerating the whole file.

Where separation still struggles, and how to check the result

Separation is a model's best estimate, not an undo button. It works best when the speech sits clearly above the music, and it degrades when the two overlap in the same frequencies: a sung vocal over instruments, a speaker talking over a loud chorus, or crowd noise that sounds like speech. In those passages you may hear faint traces of the original voice in the background, or music that dips while a line plays.

A two-minute listening test catches most problems before you spend minutes on the whole video:

• Pick the loudest music passage that has speech over it and play it twice with headphones.
• Listen underneath the new voice for a faint ghost of the source language.
• Check the joins between segments: the music should not jump in volume or restart.
• Compare overall loudness with the original — dialogue should sit above the music without the music pumping.
• If one line is the problem, re-dub that line; if the whole source is dense with sung vocals, subtitles may suit that video better than a dub.