Overlapping subtitles

Two subtitles at once, without one hiding the other.

To show two subtitles at the same time, give them overlapping times. SRT and WebVTT need nothing more, and a renderer that avoids collisions stacks them; in our ffmpeg test the line that started later sat above the other. ASS is the exception. Its Layer column turns collision checks off between layers, so two lines on different layers print on top of each other unless you give one its own position. In suubbs you drag a line onto another row of the timeline.

Get the Mac app — $9.99 → Try it free in your browser
The short answer

Overlap the times. Then check how the file draws them.

No subtitle format forbids two lines sharing a moment. What differs is who decides where the second one goes.

Most subtitles take turns: one line clears, the next appears. Sometimes two things need to be on screen together. Two people talk over each other. A sign or a caption in the picture needs translating while someone is speaking. A sound has to be described under the dialogue, which is ordinary in closed captions (see subtitles vs closed captions). Or a speaker label runs alongside the line it belongs to.

In every common format the way to do it is the same: write two cues whose time ranges overlap. The file does not need a special tag. The question is what happens when a player or a burn-in reaches the moment where both are live, and that depends on the format:

Format How the overlap is written What decides the layout
SubRip (.srt)Two cues with overlapping timestampsThe player. libass, which ffmpeg burns in with, stacked them in our test
SRT muxed into an MP4A mov_text trackThe overlap is cut: the earlier line ends when the later one starts
WebVTT (.vtt)Two cues with overlapping timestampsThe player. WebVTT's layout rules move a caption that would overlap another
Advanced SubStation (.ass)Overlapping events, each with a Layer numberSame layer: stacked. Different layers: drawn over each other
Word-level JSONsuubbs' own cues, each with its rowNothing yet. It is for re-opening the project, not for playback
1 · Before you overlap

Often the better answer is one subtitle with two lines.

Two separate cues are not the only way to show two speakers.

When two people speak in quick exchange, subtitling convention is usually a single subtitle with one line per speaker, each starting with a dash:

- Are you coming?
- In a minute.

That is one cue, so every player shows it the same way, and it cannot come apart in a burn-in. In suubbs, type the dashes into the line, then hold Alt and click the first word of the second speaker's part in the cue list to start a new line there. The cost is timing: both halves appear together and leave together, so the reader sees the answer before it is spoken.

Separate overlapping cues are the right tool when the two things have genuinely different timing, such as an interruption that starts halfway through someone's sentence, or a different purpose, such as a translated sign or a sound description that is not dialogue at all. Keep reading speed in mind either way. Two lines at once is twice the reading, and a viewer who is reading both is not watching the picture. The limits are on subtitle reading speed.

2 · In suubbs

Rows on the timeline are layers.

A line can sit on any of up to ten rows, and two rows may hold lines at the same moment.

  1. Add the second line. Put the playhead where it should start and press N, or click + Add subtitle above the cue list. A new line reading New subtitle appears at the playhead. Type over it with the real text.
  2. Drag it to its own row. Under the waveform, the timeline shows an empty row below the rows in use. Drag the line down into it. If you drop a line onto another one, suubbs moves it to the first free row rather than hiding it underneath. It also never shoves the other line out of the way.
  3. Line up the edges. While you drag, a line's start or end that comes within a few pixels of another line's start or end, on any row, is pulled onto it. Making an interruption begin exactly as the first line ends, or two lines leave together, is a drag rather than a typed timestamp.
  4. Check the picture. The preview draws every line that is live at the playhead. The one you edit by default is the line on the uppermost row of the timeline. The others are drawn a step further up the frame so they read as a stack. They are not editable where they stand. Click one in the timeline or the cue list to work on it.

Everything else works per line as usual: its own words, its own timing, its own colour if you want the second speaker or the sound description to look different. Rows only decide what may overlap.

Two lines, one moment, both readable.

[both speaking at once]

3 · In the file

What each export does with a second row.

We burned the same two overlapping lines into a test frame with ffmpeg 6.1 and libass 0.17 to see each case.

SRT and WebVTT: overlapping timestamps

Neither format has rows. suubbs writes every line in order of its start time and numbers them, so a line on the second row simply becomes a cue whose time range overlaps its neighbour's:

1
00:00:00,000 --> 00:00:04,000
First line of dialogue

2
00:00:01,000 --> 00:00:03,000
Second speaker at the same time

Burned in with ffmpeg's subtitles filter, that file showed both lines, one above the other, with the later one on top. WebVTT's rendering rules in the specification do the equivalent for browsers: a cue box that would overlap one already on screen is moved until it does not. A player that ignores those rules may still draw one line over another, so check the players your audience uses. Which line ends up on top is the renderer's choice in both formats, not yours.

One place the overlap does not survive is an MP4's own subtitle track. When we muxed that same SRT into an MP4 as a soft track (mov_text) and extracted it again, the first line now ended at one second, the moment the second began, instead of at four. If the subtitles are going into the video file rather than beside it, check them after muxing (adding an SRT to an MP4 covers the command), or keep them as a separate file.

ASS: the Layer column, and the trap in it

ASS events carry a Layer number as their first field, and suubbs writes each line's timeline row there: the first row is layer 0, the next is 1, and so on. The ASS specification says events on different layers are ignored by collision detection, and higher layers are drawn over lower ones. In practice:

So the very thing that keeps two lines apart in the suubbs preview, separate rows, puts them on top of each other in a burned-in ASS file. Until the export accounts for it, fix the file before you burn in. Either put every event back on one layer and let libass stack them:

sed -E 's/^Dialogue: [0-9]+,/Dialogue: 0,/' captions.ass > stacked.ass

or keep the layers and lift the second row with its own vertical margin. The eighth field of a Dialogue line is MarginV. suubbs writes 0 there, which means "use the style's". Any other number overrides it for that line, in pixels from the edge the caption is anchored to, which for ordinary subtitles is the bottom:

sed -E 's/^Dialogue: 1,([^,]*),([^,]*),Default,,0,0,0,/Dialogue: 1,\1,\2,Default,,0,0,130,/' captions.ass > raised.ass

Both versions came out readable in our test. The first is simpler and lets the renderer choose the order. The second gives you a fixed place for the second row, which suits a translation or a label that should always sit in the same spot. Pick a margin larger than the style's own MarginV (the second number from the end of the Style: line) by at least the height of the lower line. The same applies with word highlighting on, where suubbs writes one event per word: the events keep their line's layer, and with everything on layer 0 our test stacked the two highlighted lines and kept them in place as each word lit.

One more case. If you dragged the caption to a custom spot on the video, suubbs writes an exact \pos() on every event, and a positioned event takes no part in collision detection at all. In our test two positioned lines landed on top of each other even on the same layer. Change the second number inside \pos(…) on the lines that should sit higher. More on what is in the file is on the ASS subtitle file, and on burning it in on burning subtitles into a video.

Word-level JSON: keeps the rows

The JSON export writes each line as it is in the project, row number included, so a project exported and opened again comes back with its lines on the rows you left them on.

Questions

What people ask next.

Can an SRT file show two subtitles at the same time?

Yes. Give the two cues overlapping timestamps; nothing else is needed. How they are laid out is the player's decision. ffmpeg's subtitles filter, which uses libass, stacked them one above the other in our test, with the later-starting line on top.

Why do my overlapping ASS subtitles print on top of each other?

Because they are on different layers. The ASS specification excludes events on different layers from collision detection, so two lines that share a style position are drawn in the same place. Put both events on the same Layer number, or give one of them its own MarginV or \pos.

What does the Layer field in an ASS file do?

It is the first field of every Dialogue line. Higher layers are drawn over lower ones, and only events on the same layer are checked for collisions and moved apart. It is meant for deliberate overlaps such as a sign drawn over a background box, not for keeping dialogue apart.

How do I show two speakers in one subtitle?

Write one cue with a line per speaker, each starting with a dash. It is one subtitle, so every player shows it the same way. In suubbs, type the dashes and Alt-click the first word of the second speaker's part to break the line there.

How many subtitles can be on screen at once in suubbs?

The timeline has up to ten rows, and lines on different rows may share any moment. In practice two is a lot for a viewer to read; more than that is for special cases like translated signs over dialogue.

Will the second line keep its own colour?

Yes. Colour is set per line in suubbs and is written into the SRT, WebVTT and ASS exports. The row only decides what may overlap; it does not change the look.

Keep reading

Related

Captions that stay readable when two people talk.

[two rows, one moment]