lotobo/subtitle-tools
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases 2 Wiki Activity

Release Subtitle Tools 0.2.0
Verify / verify (push) Canceled after 0s
Details

Browse Source
This commit is contained in:
zemion
2026-09-02 08:44:49 +02:00
parent b92793eb82
commit 88c2bcd0ff
36 changed files with 1354 additions and 182 deletions
Download Patch File Download Diff File Expand all files Collapse all files
public/docs/ARCHITECTURE.md
+15 -7
View File
@@ -2,16 +2,24 @@
Subtitle Tools is a relocatable React application with a framework-independent
TypeScript subtitle model. The parser first applies byte/text and line bounds,
then produces one normalized cue model for SRT, WebVTT and ASS. Operations return
new documents instead of mutating parser state. Serializers are format-specific;
cross-format export deliberately exposes its preservation boundary.
then produces one normalized cue model for SRT, WebVTT, ASS and a bounded flat
TTML profile. XML document types/entities are rejected; TTML resources,
inherited timing and layout are never resolved. Cue structure/text/timing
operations return new documents instead of mutating parser state. Serializers
are format-specific; cross-format export deliberately exposes its preservation
boundary.
Parsing and timing transformations are synchronous in v0.1 and therefore capped
Parsing and timing transformations are synchronous in v0.2 and therefore capped
at 4,000,000 UTF-16 units, 250,000 units per physical line and 100,000 cues. The
editor renders at most 10,000 cue rows and 2,000 issue rows, while validation and
export still cover the complete bounded document. Revision comparison is
index-aligned and linear rather than an unbounded semantic diff.
The optional waveform component uses browser-native media decoding, keeps at
most 2,000 peak buckets and accepts no more than 128 MiB or four decoded hours.
It does not run a codec, speech-recognition model or network service.
The optional waveform component creates a browser-native playback URL as soon as
media is selected. Decoding continues independently and transfers one PCM
channel to a disposable module worker, which produces at most 2,000 peak
buckets. Replacement and cancellation terminate the worker, suppress stale
results and retain the last valid waveform. Zoom/navigation changes only the
bounded peak view and native playback position. Inputs remain capped at 128 MiB
and four decoded hours; no codec, speech-recognition model or network service
runs.