youtube_abschrift_korrigieren/vtt2txt.py

#!/usr/bin/env python3
"""Wandelt WebVTT-Untertitel (z. B. YouTube-Auto-Untertitel) in eine Rohabschrift um,
die anschließend mit rede_korrigieren.py bearbeitet werden kann."""
from pathlib import Path
import argparse
import glob
import html
import os
import re
import sys

TAG = re.compile(r"<[^>]*>")
GLOB_CHARS = re.compile(r"[*?\[]")


def read_cues(vtt):
    """Yield the non-empty, tag-free text lines of each cue."""
    cue = None

    for line in vtt.splitlines():
        if "-->" in line:
            if cue is not None:
                yield cue
            cue = []
            continue
        # Only a truly empty line ends a cue; YouTube puts lines with a single
        # space inside cues, followed by more text
        if not line:
            if cue is not None:
                yield cue
            cue = None
            continue
        # Header, NOTE/STYLE blocks and cue identifiers
        if cue is None:
            continue

        text = " ".join(html.unescape(TAG.sub("", line)).split())
        if text:
            cue.append(text)

    if cue is not None:
        yield cue


def overlap(previous, cue):
    """Length of the longest end of `previous` that starts `cue`."""
    for size in range(min(len(previous), len(cue)), 0, -1):
        if previous[-size:] == cue[:size]:
            return size
    return 0


def vtt_to_lines(vtt):
    """Return the caption text lines of a VTT document without rolling duplicates.

    Auto captions show a rolling window: each cue repeats the end of the previous
    cue before adding new text. Only the lines after that overlap are new, so a
    line the speaker really says twice in a row is kept.
    """
    lines = []
    previous = []

    for cue in read_cues(vtt):
        # Empty cues (YouTube transitions) say nothing about the window
        if not cue:
            continue
        lines.extend(cue[overlap(previous, cue):])
        previous = cue

    return lines


def convert(source, target):
    """Convert one file; return an error message or None on success."""
    try:
        vtt = source.read_text(encoding="utf-8-sig")
    except UnicodeDecodeError:
        return f"Keine UTF-8-Datei: {source}"

    lines = vtt_to_lines(vtt)
    if not lines:
        return f"Keine Untertitel gefunden: {source}"

    target.parent.mkdir(parents=True, exist_ok=True)
    target.write_text("\n".join(lines) + "\n", encoding="utf-8")
    print(f"Erstellt: {target} ({len(lines)} Zeilen)")
    return None


def expand_inputs(patterns):
    """Resolve glob patterns (for shells that do not, e.g. on Windows).

    Returns the files in order without duplicates, or the first pattern
    without match.
    """
    sources = []
    for pattern in patterns:
        if GLOB_CHARS.search(pattern):
            matches = sorted(glob.glob(pattern))
            if not matches:
                return None, pattern
        else:
            matches = [pattern]
        for match in matches:
            path = Path(match)
            if path not in sources:
                sources.append(path)
    return sources, None


def parse_args():
    parser = argparse.ArgumentParser(
        description="Wandelt VTT-Untertitel in eine Textdatei für rede_korrigieren.py um."
    )
    parser.add_argument(
        "-i", "--input", required=True, nargs="+",
        help="VTT-Datei(en) oder Muster, z. B. \"ramstein.de*.vtt\"",
    )
    parser.add_argument(
        "-o", "--output",
        help="Ausgabedatei (nur bei einer Eingabedatei) oder Verzeichnis "
        "(vorhanden oder mit / am Ende); Standard: <Eingabe ohne Endung>.txt",
    )
    args = parser.parse_args()

    sources, unmatched = expand_inputs(args.input)
    if unmatched is not None:
        parser.error(f"Keine Datei passt zu: {unmatched}")
    for source in sources:
        if not source.is_file():
            parser.error(f"Eingabedatei nicht gefunden: {source}")
        if source.suffix.lower() != ".vtt":
            parser.error(f"Keine VTT-Datei: {source}")

    output = args.output
    to_dir = output is not None and (
        output.endswith(("/", os.sep)) or Path(output).is_dir()
    )
    if output is not None and not to_dir and len(sources) > 1:
        parser.error("Bei mehreren Eingabedateien muss -o/--output ein Verzeichnis sein.")

    args.jobs = []
    for source in sources:
        if output is None:
            target = source.with_suffix(".txt")
        elif to_dir:
            target = Path(output) / source.with_suffix(".txt").name
        else:
            target = Path(output)
        if target.resolve() == source.resolve():
            parser.error("Ausgabedatei darf nicht die Eingabedatei sein.")
        if any(target.resolve() == other.resolve() for _, other in args.jobs):
            parser.error(f"Mehrere Eingabedateien ergeben dieselbe Ausgabedatei: {target}")
        args.jobs.append((source, target))
    return args


def main():
    args = parse_args()
    failed = 0
    # Keep going on errors so one bad file does not stop a batch
    for source, target in args.jobs:
        error = convert(source, target)
        if error:
            print(f"FEHLER: {error}", file=sys.stderr)
            failed += 1
    if failed:
        sys.exit(1)


if __name__ == "__main__":
    main()