Shorts editor splits audio from video: make voice and music files

HeyOrca reported (read 2026-10-07) that the Shorts editor now separates the audio and video tracks. The practical effect for a producer is that two clean files are more useful than one mixed file: you can nudge the voice without moving the music, and you can swap the bed without re-recording.

Sume produces both as standalone audio files, so nothing is mixed in until you decide.

What each file costs

TTS 1.0 returns a wav (pcm_s16le, 44100 Hz) and can return word timestamps, which tell you where each line lands. Music Router returns one track per generation, no vocals if you end the prompt with 'Instrumental, no vocals'. There is no duration parameter, so ask for length in the prompt, for example 'a 45-second track'.

Why keep them apart

With one mixed file, fixing a loud bed means a new render. With two files, you pull one fader. Sume's own Timeline can also duck a bed under speech (duck_db 0 to 20), at $0.10 per output minute, but that is for the case where you want one baked file. If the YouTube editor handles the mix, you can skip it.

The 15 characters per second rate is an assumption for planning. Read the real length from the word timestamps before you cut picture to it.