Guide track: audio in sync but not for delivery
A guide track is audio used to build against rather than to deliver. Treating generated speech that way is the safer assumption: the alignment is free and the levels, room tone and music are not under control. As of 2026-09-12.
| What generated speech gives you | What it usually does not | Why it matters |
|---|---|---|
| Alignment with the picture | A consistent level between shots | A mix has to be rebuilt anyway |
| A reading of the line | A direction you asked for | Nothing here takes performance notes |
| A bed of effects and music | A stem you can replace | Only one entry documents isolating speech |
Inclusion rule. The properties generated speech reliably has, against the properties a delivery mix needs, as documented across the entries in this register. Order. Fixed order: alignment, performance, then the mix.
1Alignment is the expensive problem, and it comes free
Where speech is generated with the picture, mouth and waveform agree by construction. That is the part a dubbing pipeline spends money on, and it arrives at no cost on every entry that generates sound.
Everything else about the audio is less settled. Levels move between calls, room tone changes, and music comes and goes with the prompt, so a sequence assembled from several generations rarely sounds like one scene.
2Planning for replacement keeps an estimate honest
If a production assumes generated dialogue is a guide track, then a session to replace it is in the budget and a pleasant surprise is possible. If it assumes delivery audio, the session appears as an overrun.
The asymmetry is the argument. One assumption can only be beaten and the other can only be missed, and neither is a statement about how good any model sounds.
3A guide track has one hard requirement
For generated speech to be usable as a guide, the words have to be right even if the reading is not. That makes the dialogue syntax the part worth getting exactly correct, because a misheard word propagates into a replacement session as well.
It also means a shot can be approved on picture while its audio is still provisional, which is only possible where redoing the sound does not redo the picture. Most entries here do not allow that separation.
4Sources read for this entry
This page defines the term and logs what each vendor documents about it, each figure read on 2026-09-12. Related: Dubbing pass, Timbre.