Split WebVTT cues that use cue timestamps (https://www.w3.org/TR/webvtt1/#webvtt-cue-timestamp) into multiple cues for VLC

webvtt_normalize_timestamps.py

# SPDX-License-Identifier: CC0-1.0

# Splits WebVTT cues that use cue timestamps (https://www.w3.org/TR/webvtt1/#webvtt-cue-timestamp) into multiple cues
# This is mainly relevant for auto-generated YouTube captions, which generally aren't split on sentence boundaries and thus showing the whole cue produces bad results

import webvtt  # https://github.com/glut23/webvtt-py/ - pip install webvtt-py
import sys
import os

if len(sys.argv) > 2:
    in_file = sys.argv[1]
    out_file = sys.argv[2]
else:
    out_file = sys.argv[1]
    in_file = out_file + '.old'
    os.rename(out_file, in_file)

caption_data = webvtt.read(in_file)
old_captions = list(caption_data.captions)
caption_data.captions.clear()

for caption in old_captions:
    if '<' in caption.raw_text:
        new_caption = ''
        start = caption.start
        for part in caption.raw_text.split('<'):
            if '>' in part:
                tag_data, rest = part.split('>')
                if webvtt.models.Timestamp.PATTERN.fullmatch(tag_data):
                    caption_data.captions.append(webvtt.models.Caption(start, tag_data, new_caption, caption.identifier))
                    start = tag_data
                    new_caption += rest
                else:
                    new_caption += '<'
                    new_caption += tag_data
                    new_caption += '>'
                    new_caption += rest
            else:
                new_caption += part
        caption_data.captions.append(webvtt.models.Caption(start, caption.end, new_caption, caption.identifier))
    else:
        caption_data.captions.append(caption)

caption_data.save(out_file)