Skip to main content

Crate rightkit_transcript

Crate rightkit_transcript 

Source
Expand description

rightkit-transcript: the shared word-timed transcript contract.

Pipeline: producer (ASR, TTS manifest, SRT/VTT file) -> Transcript -> build_cues (CaptionConfig) -> to_srt / to_vtt / timing JSON.

  • contract: rightkit.transcript.v1 (reads cutright.transcript.v1).
  • words: TimedWord normalisation and source -> output time mapping.
  • segment: deterministic cue segmentation with named profiles, plus the lightweight group_words grouper for already-clean engine words.
  • export: SRT, WebVTT, plain text, and the shared timing JSON.
  • parse: SRT/WebVTT/timing-JSON import (cues -> interpolated words).
  • align: cue-edge vs word-boundary check used by qualifiers.

All times are integer milliseconds. No audio, no models, no I/O here: engines hand in (text, start, end) and apps write the strings out.

Re-exports§

pub use align::check_alignment;
pub use align::AlignmentIssue;
pub use align::ALIGNMENT_TOLERANCE_MS;
pub use contract::EndSource;
pub use contract::ModelRef;
pub use contract::Span;
pub use contract::Transcript;
pub use contract::TranscriptError;
pub use contract::Word;
pub use contract::LEGACY_SCHEMA;
pub use contract::SCHEMA;
pub use export::to_plain_text;
pub use export::to_srt;
pub use export::to_timing_json;
pub use export::to_vtt;
pub use export::Newline;
pub use parse::parse_srt;
pub use parse::parse_timing_json;
pub use parse::parse_vtt;
pub use parse::words_from_cues;
pub use parse::ParseError;
pub use parse::ParsedCue;
pub use segment::build_cues;
pub use segment::cues_from_spans;
pub use segment::group_words;
pub use segment::normalize_words;
pub use segment::CaptionConfig;
pub use segment::Cue;
pub use segment::GroupRules;
pub use segment::Profile;
pub use words::words_from_seconds;
pub use words::words_from_transcript;
pub use words::TimedWord;
pub use words::MAP_DURATION_TOLERANCE_MS;

Modules§

align
Word-boundary alignment check (used by qualifiers).
contract
rightkit.transcript.v1: word-timed speech for one source.
export
SRT, WebVTT, plain-text and timing-JSON export. UTF-8, exact millisecond timestamps.
parse
SRT / WebVTT / timing-JSON import. Cues carry no word timings, so words_from_cues spreads each cue’s words across its span by character count (flagged EndSource::Interpolated when put into a Transcript).
segment
Cue segmentation: phrase grouping, splitting, merging, line breaking.
words
Word timing input: TimedWord in OUTPUT time, plus engine adapters.