Toolvore

Subtitle Quality Checker

Check plain SRT or WebVTT subtitles against your own reading-rate, line-length, duration and gap limits, then export a cue-by-cue issue report.

This tool processes its input contents in your browser instead of sending them to a processing service. Page, asset and fixed usage-measurement requests are described on the Privacy page.

Check plain SRT/WebVTT cue reading rate, length, duration, gaps and overlaps against explicit settings, with a downloadable issue CSV and settings report.

How to use it

  1. 1Paste subtitle text or load a UTF-8 SRT/WebVTT file, and select its format.
  2. 2Set the recipient's actual limits; optional duration, gap and line-count controls are available below.
  3. 3Check, review the cue findings and download the issue CSV and JSON summary.

Example

Input
1 00:00:00,000 --> 00:00:01,000 Hello! 2 00:00:00,900 --> 00:00:02,900 Next cue.
Output
cue,source_line,issue,observed,limit,related_cue 2,5,overlap,100,0,1

Default review settings find a 100 ms overlap before cue 2. The original subtitle is not changed.

What happens to your data

Subtitle parsing and quality measurements run in this tab without uploading dialogue, opening media or saving history. The issue/count exports omit cue text and identifiers. Optional fixed workflow measurements exclude the subtitle contents.

Last updated November 2026

Review an existing subtitle track before you deliver it. Paste plain SRT or WebVTT, choose your reading-rate and line-length limits, then check the cues. A downloadable CSV identifies cue numbers, source lines and measured issues, while a separate JSON report records the settings used. No video upload or subtitle service is needed.

How it works

The checker measures what the supplied text and timestamps actually contain. Characters per second equals the cue's Unicode code-point count divided by its duration. Spaces and punctuation count; newline separators do not. For example, an emoji represented by one Unicode code point counts once, while a letter plus a separate combining mark counts twice. These are not grapheme-cluster counts, font widths or visual screen columns. Supported WebVTT named character references are decoded before counting.

The initial settings are 20 characters per second, 42 per line, at most two lines, durations from 1,000 to 7,000 milliseconds and a minimum 80 millisecond non-overlapping gap. They are editable review settings, not a promise that every language, audience or delivery platform uses those requirements. Only values strictly outside your limits are findings: exactly 20 characters in one second is within a limit of 20. Duration and gap controls use whole milliseconds. A shorter gap can be intentional; review it in context rather than automatically inserting silence.

Overlap is checked against the latest ending earlier cue, including a long cue that spans several subsequent cues. To keep the report useful and bounded, each affected cue gets one temporal finding referring to that earlier cue, rather than listing every overlapping pair. The checker does not reorder, shorten or repair captions. Line-length rows point to the actual payload line; other findings point to the beginning of the cue block.

SRT accepts positive numeric counters, a start/end timestamp line and nonempty cue text separated by blank lines. WebVTT must begin with WEBVTT and a blank header separator. Optional WebVTT cue identifiers, NOTE blocks and the existing parser's supported basic positioning settings are accepted; notes are not measured and settings are not rendered. STYLE, REGION, unknown settings, inline markup and embedded timing tags are outside this plain-text workflow and stop the operation. The supported named references are amp, lt, gt, nbsp, lrm and rlm; references that decode to angle-bracket markup are refused. This is a documented subset, not a complete WebVTT conformance validator.

UTF-8 input may have an initial BOM and LF, CRLF or CR line endings. Invalid UTF-8 files, unsupported control characters, unpaired surrogates, invalid timing, decreasing cue starts and empty cues fail clearly. Input is bounded to 2 MiB, 60,000 physical lines and 10,000 cues; exports are bounded to 20,000 findings and 4 MiB. The complete operation fails when a bound is exceeded rather than returning an apparently complete partial report. A header-only issue CSV means no findings under the selected settings.

Reports contain cue ordinals, source line numbers, issue codes, measurements, thresholds and settings. They omit dialogue and cue identifiers, but their structure can still reveal information about a track. Work stays in browser memory; changing the input or settings clears the previous result, and a late file read cannot overwrite a newer edit. Clear cannot erase downloaded copies. Check the original track in its actual player and against the actual media: these arithmetic checks do not verify the transcript, translation, speaker labels, audio synchronization, accessibility or any platform's acceptance.

Worked example: Find an overlapping cue

Input
1 00:00:00,000 --> 00:00:01,000 Hello! 2 00:00:00,900 --> 00:00:02,900 Next cue.
Result
cue,source_line,issue,observed,limit,related_cue 2,5,overlap,100,0,1

Default review settings find a 100 ms overlap before cue 2. The original subtitle is not changed.

Common use cases

  • Create a repeatable handoff checklist using the recipient's actual subtitle limits.
  • Find rapid captions, long lines and duration outliers in a small SRT export.
  • Locate timing overlaps, including nested overlaps, without rewriting intentional cues.
  • Export a review report without copying dialogue into the report.

Frequently asked questions

Does a clean report mean the subtitles are correct or accessible?+

No. It means the supported text passed your selected arithmetic checks. You still need to review transcript accuracy, language, context, intended overlaps, audio synchronization and the recipient's requirements.

What counts toward characters per second?+

Unicode code points in the cue text, including spaces and punctuation. Line separators are excluded. Supported WebVTT named references are decoded first; graphemes and displayed widths are not measured.

Are the defaults Netflix or another platform's certificate?+

No. Defaults are editable starting points. Set the actual language, audience and delivery requirements you have been given.

Does it repair timing or remove overlap?+

No. It exports findings and leaves the original track unchanged. Simultaneous speakers and other intentional overlaps need an editorial decision.

Can an SRT or WebVTT file valid elsewhere be rejected?+

Yes. This plain-text subset rejects markup, STYLE/REGION and unsupported constructs instead of making misleading text measurements.