Eyevinn / moqlivemock / 36583253807

56%

Build Type

push

github

Committed by tobbee

Commit Message

feat(subtitles): send subtitles at the video cadence, with paint-model and LOCMAF tracks

A subtitle group held one object per second, sent at the start of the second. It now holds one CMAF chunk per video object (25 per second for the test content), each sent when the interval it covers ends, so text reaches the player with the latency of the picture it belongs to. The chunk bounds come from the first video track, computed as for its own objects.

The cues follow livesim2's chunked generator: a cue keeps its true begin in every chunk and gets its end only in the chunk where it ends, so an unchanged cue restates byte for byte and the restatement is marked redundant (sample_has_redundancy). Cue ids are c<UTC second>, and WVTT cues carry a vsid box with the same second, so a restated cue is recognisably the same one.

-subsstpc and -subswvtc add the paint-model variants from github.com/Eyevinn/paint-model-subtitles, off by default since their 4CCs are unregistered placeholders. A chunk that restates the previous one becomes an 8-byte ttmn or vttn box, and with -subsstpcbody (on by default) a changed STPC chunk after the first of its group carries only the TTML body, in a ttmb box. Both are non-sync samples that depend on the first chunk of the group.

Every subtitle track is also published as <name>_locmaf, sharing the CMAF track's init data, and the catalog gives each track's bitrate as measured on a generated group. At 25 fps that is, CMAF / LOCMAF: stpp 317 / 296 kbit/s, stpc 38 / 17, wvtt 35 / 14 and wvtc 25 / 4.5. Each subtitle format has its own altGroup.

Coverage Stats

270 of 324 new or added lines in 5 files covered. (83.33%)

7 existing lines in 2 files now uncovered.

3294 of 6055 relevant lines covered (54.4%)

1266.0 hits per line