What SBV actually is: an anatomy of YouTube's caption format

SBV is a caption format built for simplicity, and it is the format YouTube's caption tooling has historically used. It differs from SRT in four structural ways, all of them visible in a side-by-side comparison:

SRT:
1
00:00:01,000 --> 00:00:04,000
Welcome to the show.

2
00:00:05,500 --> 00:00:09,200
Today we're talking about subtitles.

SBV:
0:00:01.000,0:00:04.000
Welcome to the show.

0:00:05.500,0:00:09.200
Today we're talking about subtitles.

The four differences, one at a time

1. No cue numbers

SRT begins each cue with a line number. SBV has no numbering at all — the file is just timestamps and text separated by blank lines. Converting SRT to SBV discards those 1, 2, 3 lines entirely. The parser does read them, but the SBV serialiser has nowhere to put them.

2. Both times on one line, joined by a comma

SRT puts the two times on their own line joined by an arrow: 00:00:01,000 --> 00:00:04,000. SBV joins them with a plain comma and no arrow: 0:00:01.000,0:00:04.000.

3. A dot before the milliseconds, not a comma

SRT writes 00:00:01,500; SBV writes 0:00:01.500. This is not arbitrary. Since the comma is already doing the job of separating start from end, a comma before the milliseconds would give a single line four comma-delimited pieces with no way to tell which is which. Using a dot keeps the line unambiguous without any escaping or quoting — the format is internally consistent by design.

4. A single-digit, unpadded hour

SRT pads the hour to two digits. SBV does not, so one hour and change reads as 1:05:30.000 rather than 01:05:30,000. This tool emits exactly H:MM:SS.mmm. Note this is independent of the separator question above — it is simply a different convention, and a parser expecting a zero-padded hour will reject valid SBV.

What stays the same, and what the tool does with damage

Two things carry across unchanged. Cues are separated by blank lines in both formats, and multi-line cue text is preserved exactly, with its inner newlines intact. Timing is exact too: times are held as integer milliseconds and only reformatted, so no precision is lost.

Separately from any format difference, the converter is tolerant of damaged input. A cue block whose timestamps cannot be parsed is skipped and named by block number in a warning, and everything else converts normally — one corrupt line in a 400-cue file will not cost you the other 399. Only a file with no usable cues at all is a hard error.

SBV is specific to one platform's uploader, so it is worth being sure that is the destination before converting. For anything web-based, SRT to VTT is the more broadly useful conversion.