Subtitle release matching
Subtitle release matching compares identity, cut, timing and optional file fingerprints to choose a candidate suited to the actual media edition.
A subtitle must match the release being played, not merely the title listed in the library. Different cuts, opening material, recaps, frame playback speeds and episode combinations can change the content timeline.
Language is another independent requirement. A perfectly synchronized track in an unsuitable script or role can still be the wrong viewing choice.
Combine evidence rather than one label
| Clue | What it helps establish |
|---|---|
| Title/year or series/episode identity | The intended work |
| Edition and release description | A proposed cut/source relationship |
| Duration and recognizable scene positions | Timing correspondence |
| File fingerprint match | Association with a particular byte-based release identity |
| Language, forced/SDH and content metadata | The intended reading content |
| Several verified cue positions | Actual synchronization across the programme |
A filename's release-group text is an identification clue, not proof. So is a duration: two different edits can have nearly equal total lengths.
Compare distinct scenes. A repeated generic sentence is a poor anchor because it can match the wrong moment.
OSHash is a sampled file fingerprint
The published OpenSubtitles algorithm adds the file size and little-endian 64-bit words from the first and last 64 KiB of a file, keeping the result modulo 2⁶⁴. The reference specifies the applicable file-size requirements and exact block handling.
It is not a full-file cryptographic digest. Its matching role should be described accordingly.
For a small arithmetic illustration—not a complete OSHash calculation—take a size term of 200,000, one word of 2⁶⁴ − 16, and another word of 48:
(200000 + (2^64 − 16) + 48) mod 2^64
= 200032
= hexadecimal 0000000000030d60
The example shows unsigned wrap/sum behavior. A real fingerprint uses all the required sampled words; replacing it with this three-term sum would not implement the algorithm.
File identity and timing identity differ
A remux or transcode can change the file's bytes while retaining the same programme timing. A new fingerprint therefore does not automatically require retiming a subtitle.
Conversely, an extra 60-second scene can change all later cue positions while earlier ones remain aligned. Edition mismatch explains that piecewise timing relationship; a global duration stretch can damage the earlier section.
For an Emby candidate comparison, keep the actual media edition, language/role and sampled cue evidence together. Provider match metadata can prioritize the search, while checking the resulting presentation closes the match to the movie being watched.