Fixing Garbled Characters in SRT and VTT Caption Files
A broadcaster rejected a 44 minute subtitle file I delivered three years ago because the Norwegian director's surname came back with a capital A and a circumflex wedged in front of every accented letter, symbols that belong in neither...

A broadcaster rejected a 44 minute subtitle file I delivered three years ago because the Norwegian director's surname came back with a capital A and a circumflex wedged in front of every accented letter, symbols that belong in neither language. I had checked the file in my subtitling software, where it looked perfect. The ingest system opened it with a different assumption about what the bytes meant, and that one disagreement cost me a day and a redelivery fee.
Garbled characters in caption files are almost never a typing mistake, and they are almost never fixed by retyping the affected words. They are a character encoding mismatch: the program that wrote the file and the program that read it disagreed about which table of characters the raw bytes point to. Once you can read the symptom, the fix usually takes under a minute.
Reading the symptom tells you the cause
The specific shape of the corruption is diagnostic. I keep this table taped inside a notebook because the pattern repeats across every platform I have delivered to.
| What you see | What happened | Fix |
|---|---|---|
| Accented letters become two characters, so cafe gains a capital A with a circumflex before the accent | UTF-8 bytes read as Windows-1252 | Reopen declaring UTF-8, resave. The file itself is fine. |
| Accented letters become a single question mark or a black diamond | Windows-1252 or Latin-1 bytes read as UTF-8, data already lost on read | Reopen the original declaring Windows-1252, then save as UTF-8 |
| A stray character before the first subtitle number, or player reports "invalid index" | UTF-8 byte order mark at the start of the file | Save without BOM. The three bytes are EF BB BF. |
| Cyrillic, Greek or Thai text replaced entirely by Latin gibberish | File saved in a legacy regional code page | Identify the code page with a converter, transcode once |
| Every character separated by a space or a null in a hex viewer | Saved as UTF-16, which most caption parsers reject | Convert to UTF-8 without BOM |
| Curly apostrophes appear as a capital A with a tilde plus a symbol | Smart punctuation written in UTF-8, read as single byte | Transcode, or replace smart punctuation with straight marks before export |
The second row is the dangerous one. When Latin-1 bytes get read as UTF-8 and the application substitutes replacement characters, the original letter is gone. No later conversion brings it back, because the information was discarded at read time. If that file is the only copy you have, you are retyping names.
The advice that sounds right and is not
Search for this problem and the answer you will find is "always save your subtitles as UTF-8". I have given that advice myself and I have watched it fail on delivery.
Two things break it. First, a plain UTF-8 declaration says nothing about the byte order mark, and the BOM is what trips strict parsers. A WebVTT file tolerates it because the specification says to skip it. Plenty of SRT parsers, including several set-top box firmwares and at least one of the two major ingest tools I work with, treat those three bytes as part of the first cue number and fail the file. So the instruction has to be UTF-8 without BOM, not UTF-8.
Second, not every destination wants Unicode at all. Broadcast subtitle formats carry their own character set rules. An EBU STL file uses a code page declared in its own header, and sending it something outside that set produces blanks rather than accents. A few archive systems I have delivered into still specify Latin-1 explicitly in their spec sheet. The rule that actually holds is narrower and duller: read the destination's spec, encode to that, and verify after conversion rather than before.
The related bad advice is to fix the visible characters by hand. On a 600 cue episode you will miss some, and you will also leave the underlying encoding wrong, so the next person who touches the file starts the cycle again. Fix the encoding, then spot check the names.
Where the corruption gets introduced
In my experience four steps account for nearly all of it.
Notepad on Windows is the most frequent culprit. Older builds default the Save As encoding to ANSI, which is Windows-1252 on a Western install, and a subtitler who opens a UTF-8 file to fix one typo and hits save has quietly transcoded the whole file. The newer Notepad defaults to UTF-8, but the dialogue still remembers whatever you last chose, and I have been caught by that.
Spreadsheet round trips come second. Exporting cues to CSV to send a client a text pass, then reimporting, passes the text through the spreadsheet's own encoding choice, which on a Mac and on a PC are not the same. If you must do this, export and import explicitly as UTF-8 at both ends and compare a known accented line.
Automatic speech recognition output is a quieter source. Several transcription services return text with typographic apostrophes and ellipsis characters that live outside the single byte sets. The transcript looks clean, then one leg of your pipeline downgrades it. Our note on fixing auto generated captions covers the editing passes; add a find and replace for smart punctuation to that list if your destination is a legacy system.
Finally, email and chat clients. A caption file pasted into a message body rather than attached arrives re-encoded and often with the line endings changed too. Always attach, and zip it if the client's mail filter strips sidecar files.
A verification routine that takes two minutes
Before any delivery I run the same short check, and it has caught three files this year that looked fine on screen.
Open the file in a text editor that shows and lets you set the encoding, VS Code and Notepad++ both do, and confirm the status bar says what you expect. Then search the file for the accented or non-Latin characters you know are in it. I keep a short string of test characters in the notes field of every project: one acute accent, one umlaut, one eszett, one curly apostrophe. If all four render, the file is consistent.
Next, check the first byte. In Notepad++ the encoding menu distinguishes UTF-8 from UTF-8 BOM outright. On the command line, a hex dump of the first sixteen bytes tells you in one glance, since an SRT should begin with the digit 1 and a line break.
Then load the file into a player that is not the software you authored in. VLC with the subtitle text encoding setting left on automatic is a reasonable proxy for how a dumb parser will behave. If VLC needs you to override the encoding manually, the destination will probably need the same thing, and that means the file is wrong rather than the player.
Keep the clean master. I store the authored file plus a copy transcoded to each destination's spec, with the encoding written into the file name, because a file called episode04_en_utf8nobom.srt removes an entire category of guessing six months later. The same reasoning applies to the format choice itself, which is covered in choosing a caption file format that survives re-uploads.
Translated files deserve extra care
Everything above gets worse the further the language sits from the Latin alphabet. A Polish file has enough accented characters that corruption is obvious within two cues. A French file can look almost readable with half its accents mangled, which is how broken files reach air.
When I take delivery of a translated subtitle file from a vendor I ask for two things: the encoding stated in the handover note, and one cue number containing a diacritic that I can check immediately. That second request has saved more time than any tool, and it fits naturally into the brief if you are already thinking about translating subtitles as a timed job rather than a text job.
FAQ
Should SRT files be UTF-8 or ANSI?
UTF-8 without a byte order mark suits almost every modern destination, and it is what I deliver unless a spec sheet says otherwise. ANSI, meaning Windows-1252 in practice, cannot represent characters outside Western European languages at all, so it fails the moment a name needs a character it does not hold.
Does WebVTT handle encoding differently from SRT?
The WebVTT specification requires UTF-8 and tells parsers to skip a byte order mark if one is present, so the format removes the ambiguity that SRT leaves open. That is one practical argument for converting to VTT for web delivery even when the platform accepts both.
Can I recover a caption file that already shows question marks?
Not from that file. Replacement characters mean the original bytes were discarded when the file was read, so the accented letters no longer exist in it. Go back to the version before the bad save, or to the authoring project, and re-export.
Why do captions look right in my editor but wrong after upload?
Your editor is remembering the encoding it used to write the file, while the upload parser guesses from the bytes alone. Test by reopening the saved file in a different application rather than by looking at your timeline, since the timeline shows you the text in memory and not the text on disk.
More in Captions and Subtitles
Captions and Subtitles
How to Add Captions to a Video in Premiere Pro
You hit Transcribe Sequence, waited four minutes, clicked Create Captions, and the text appeared over your video looking...
Captions and Subtitles
Captions for Vertical Video That Stay Readable on Phones
Your captions looked fine in the editor, then you posted the clip and the like button, the username and the description text...
Captions and Subtitles
How to Make an SRT File for Your Video Captions
Your video is finished, the platform wants a caption upload, and the auto captions it generated spell your guest's name three...