04 Subtitle & Dubbing Editor
Expert Video Translation stops as soon as it reaches the "Translation" step and drops you in here automatically; the other project types arrive after the run finishes (Quick Video Translation lands on the preview page first — click "Edit" to get here, see 03 Translation Pipeline & Progress). The layout is full-screen: the parameter panel on the left, the Video Preview in the middle, and one subtitle table across the whole bottom. There is no "Save" button — the moment the mouse leaves a cell your change is already written down. The real commit happens when you click "Generate Video".
The Subtitle Editor button at the top of the pipeline page also gets you in. In the left panel, besides Subtitle Style, there is Target Voice (Voice Clone / Built-in Characters / Custom Role) and Subtitle Language (None / Target only / Source only / Bilingual). Neither group shows up in the Expert Video Translation order form, and the editor is the only place you can set them — so on your first Expert project, don't rush to click "Generate Video": settle the voice and the subtitles in the left panel first.
1. Every column of the subtitle table
| Column | Editable | Notes |
|---|---|---|
| Timestamp | Read-only | Each subtitle shows only its start time, never the start-to-end range |
| Speaker | Editable | Only present when Multi-speaker mode is on; one click switches the speaker |
| Emotion | Editable | This entry only appears when Speech Synthesis Quality is set to its highest tier and the target language is Chinese, English or Cantonese |
| Source | Editable | The original speech as recognized. Editing it rewrites the line, so it gets re-dubbed |
| Actions | — | Expert type only: Merge with Previous, Split into Two |
| Translation | Editable | The text that actually gets spoken; the Preview Audio button sits right beside it |
| Speed | Read-only | How much this line was compressed — see section 3 below |
Cells you have edited are marked red, so you can see afterwards which ones you touched. Column widths are draggable, and the column headers stay pinned while the table scrolls.
2. How Preview Audio works
There is only one Preview Audio button, and no separate regenerate button: if the line already has dubbing it plays it, if not it generates one on the spot. Two things to know:
- Editing any subtitle clears the dubbing for the whole file. So the first Preview Audio after an edit often takes a while (it is synthesizing right then), and clicking it again gives "Speech is being generated, please wait a moment".
- Preview Audio plays at natural pace; the finished video does not. To squeeze every sentence into its time window, the finished video speeds up the lines that don't fit; Preview Audio deliberately leaves that out and the Speed column reads 1.0x. That means "it sounded unhurried in the preview" is not evidence the line fits — to judge how rushed it will be, read the Speed column, not the preview.
3. What the Speed column is telling you
It shows the factor this subtitle is compressed to relative to natural pace (1.2x, for instance). The rule: the whole video runs on one base pace, only the lines whose time window is too small are sped up individually, every other line keeps the same value; the factor only goes up (never below 1.0x) and it has a ceiling — past the ceiling a line is pushed to that limit, and you will hear that one sentence clearly faster. There is only one way to slow it down: make that translation shorter.
4. When you have to re-run, and for how long
When you click "Generate Video", the program re-runs everything from the dubbing step to the end (regenerate audio → synthesize → mux → burn subtitles). That means:
- Changing the translation, the source, or the punctuation, and merging or splitting a line — every one of these re-dubs the whole file, including the sentences you never touched, and the quota this part consumes is recalculated.
- Font, size, position, background, volume and voice go through the same button and re-run to the end just the same.
- What never needs a re-run: Preview Audio, Search, collapsing the preview, undo and redo.
- The Video Subtitle type only re-runs subtitle burning, which is far faster.
So the right way to work is: get the text completely final, then generate once — rather than generating after every single edit.
5. Undo, discard, and the other entries
- Where the source sits: the Source Beside / Source Below toggle in the toolbar moves the Source column ("Below" is easier on the eyes in a narrow window); if the video preview has been switched off, Restore calls it back.
- Undo / Redo: toolbar buttons, or
Ctrl/Cmd + ZandCtrl/Cmd + Shift + Z. Text edits rewind many steps; structural operations such as merging and splitting keep only a limited number of snapshots. - Discard Changes: returns the subtitles and the settings to the state of the last generation.
- Re-translate All: Expert type only, and you can't click it back to back (rate limited).
- Import SRT: Video Dub type only.
- Open Project Folder: the finished
target.mp4already sits in the project directory — there is no separate "export" button inside the app.
Source Language, Target Language, Speech Recognition Quality and AI Subtitle Correction are read-only in the editor — they decide the recognition result, so changing one would mean starting over from the recognition step, which is why you choose them when creating the project. To change them, go back to the project list and click New Project to start another project.
Too much merging and splitting leaves the number of translation lines out of step with the source, and the re-run reports this. Line the two sides up by hand (delete the extra lines, add a line where one is missing); you don't have to rebuild the project.