This is some code I've been using to extract subtitles into CSV pairs for the purposes of language flashcards. The target encoding is hardcoded to iso-8859-1 and it requires both subtitle files to use the same timings, but otherwise it works just fine. Modified to handle multi-line sentences better. It drops some admittedly, but the flashcard qu…

extract_subtitles.py

#!/usr/bin/python
import unicodecsv
import argparse
import sys 
import re
from pysrt import SubRipFile

# Determine which languages to import
parser = argparse.ArgumentParser()
parser.add_argument("source", help="source language file to parse")
parser.add_argument("target", help="target language file to parse")
args = parser.parse_args()

source = SubRipFile.open(args.source)
target = SubRipFile.open(args.target, encoding="iso-8859-1")
writer = unicodecsv.writer(sys.stdout, encoding="utf-8")

source_phrases = []
target_phrases = []
for entry in source:

    source_phrase = entry.text
    target_phrase = target.slice(ends_after=entry.start, starts_before=entry.end)[0].text
    if re.search("^[a-z]", source_phrase) or (source_phrases and source_phrases[-1][-1][-1] == ","):
        source_phrases[-1].append(source_phrase)
        target_phrases[-1].append(target_phrase)
    else:
        source_phrases.append([source_phrase])
        target_phrases.append([target_phrase])


for index in xrange(len(source_phrases)):
    source_phrase = source_phrases[index]
    target_phrase = target_phrases[index]
    if re.search("^[A-Z]", source_phrase[0]) and re.search("[\?\.\!]$", source_phrase[-1]):
        writer.writerow((
            " ".join(source_phrase),
            " ".join(target_phrase),
        ))