open-standards/wai/live-conformance/captions/README.md

live-conformance/captions — timed text and live captions

The corpus of extensions/timed-text.md §13: live captions as WEV1 events with their WTX1 annotations, cues given as lists, and what each maps onto: WebVTT documents and segments, IMSC1 documents, the two MoQ caption timelines, CTA-608 byte pairs, and the disclosure label in HLS, C2PA, the catalogs and the sidecar.

Timed-text equivalence. For each event vector, a conforming verifier of the corpus's room, holding every principal to live-events' default policy and offered events one at a time in the order given, reaches the outcome the vector states (§4.3): refuse with the reason and index, gap with the gaps, or accept. A vector that does not refuse also states its revisions, what a sink shows at each instant of shown (§4.4), and the cues the stream resolves to or the refusal (resolve, §4.5); a vector with a join states the records a subscriber joining the events track at that group receives, its outcome and what it shows (§4.6). For each event vector that resolves and each cue vector, targets states, for each target it names, the file written or the reason refused: webvtt (§5.1), segments (§5.2), imsc (§6.1, §6.2), timelines (§7.1–§7.3; every object byte for byte), cc608 (§8; the mode, the slot clock's rate, the file and what the decoder reports of it, §8.8) and disclosure (§7.4, §10: the HLS tags, the session data, the sidecar, the C2PA statement and the catalog entries, each byte for byte). read_cases are WebVTT texts the reader refuses (§5.3), and decode_cases 608 streams the decoder refuses (§8.8), each with its reason. A verifier that reaches another outcome, reason, index, revision, cue, rendering or byte does not conform.

Provenance. Generated here: wai_timed_text_vectors gen writes vectors.json and every expected output from events and cues authored in code, from fixed seeds. Each principal's key and each model digest is SHA-256 of a fixed label and an index, the room id likewise, every time is a constant, and Ed25519 signing is deterministic (RFC 8032), so every file is the same bytes on every machine. The texts are written for the corpus, in English and Spanish. The generator asserts, before it writes, that each vector reaches the outcome it was written to have, that every WebVTT file reads back to its cues, that every 608 file decodes to the rows its cues lay out, and that the corpus covers every reason of §12 and every 608 code path of §8. No recorded data.

Layout

• vectors.json: pt0_ms (the media time of the room's t0), room, principals (captioner, asr, host), models, event_vectors, cue_vectors, read_cases and decode_cases. A cue is written as start, end, lang, origin, model (lowercase hexadecimal, or null), anchor, text and italic (pairs of first code point and count).
• NAME.vtt, NAME.segK.vtt: WebVTT documents and segments.
• NAME.ttml: IMSC1 Text Profile documents.
• NAME.MODE[.rNUM-DEN].608.hex: 608 byte pairs in the hexadecimal form of §8.3, on the 30000/1001 slot clock unless the name gives another rate.

Checking

cargo run --no-default-features --features timed_text --bin wai_timed_text_vectors -- verify ../live-conformance/captions
python3 -I timed_text_verify.py captions

The first runs from wai-rs/, the second from live-conformance/. The second is the corpus's second implementation, written from the extension's text; it shares no code with wai-rs.