Dubbing a video with music: what replacing the audio track means

When you dub a video with Sume's separate steps, you transcribe the speech, translate it, speak it with TTS and place the new voiceover on a Timeline render. The new voiceover becomes the audio spine. Anything that was in the original track, including music and room sound, is not carried over automatically.

What the pieces do

• Audio detach pulls the audio of one video into a wav, for 1 cent per job.
• STT returns the words and times at $0.01 per audio minute.
• TTS speaks the translated lines at $0.0475 per 1,000 characters.
• Timeline places the voiceover and, optionally, a soundtrack bed under it.

About separating music from speech

I did not find a stem-splitting or music-removal job in Sume's model docs, so plan as if one track is all you get from the source. That leaves three honest options.

• If you have the original music or sound mix as its own file, import it and use it as the Timeline soundtrack.
• If you do not, generate or choose a new bed and set duck_db (0 to 20) so it dips under the voice.
• If the original sound matters, such as a live event, dub only the talking sections and keep the source audio for the rest.

Settings for the bed

The documented soundtrack fields are url, gain_db, loop, fade_out_seconds (up to 10) and duck_db. Ducking needs a real voice spine, which a dub always has. Start with gain_db around -18 and duck_db 10, listen, and move in 3 dB steps.

A cost sketch for a 3-minute video

Detach: $0.01 per job, up to 900 seconds each. STT: 3 minutes is $0.03. TTS: 3 minutes of speech is roughly 2,700 characters, about $0.13 per language (billed in whole cents per job). Timeline render fees are separate. The translation step is outside Sume's list above, so add your own translator's cost.