463 lines
15 KiB
Python
463 lines
15 KiB
Python
#!/usr/bin/env python3
|
|
|
|
import re, json
|
|
import os
|
|
from datetime import timedelta
|
|
|
|
|
|
def from_vtt(vtt_string):
|
|
VTT_TIMECODE_PATTERN = (
|
|
r"((?:\d{2}:)?\d{2}:\d{2}\.\d{3}) --> ((?:\d{2}:)?\d{2}:\d{2}\.\d{3})"
|
|
)
|
|
VTT_LINE_NUMBER_PATTERN = r"^\d+$"
|
|
parts = re.split(r"\n\n+", vtt_string.strip())
|
|
if parts[0].startswith("WEBVTT"):
|
|
parts.pop(0)
|
|
|
|
subtitles = []
|
|
for part in parts:
|
|
lines = part.split("\n")
|
|
match = re.match(VTT_TIMECODE_PATTERN, lines[0])
|
|
if not match:
|
|
if re.match(VTT_LINE_NUMBER_PATTERN, lines[0]):
|
|
lines.pop(0)
|
|
match = re.match(VTT_TIMECODE_PATTERN, lines[0])
|
|
if not match:
|
|
continue
|
|
|
|
start, end = match.groups()
|
|
content = "\n".join(lines[1:]) + "\n"
|
|
subtitles.append(
|
|
{
|
|
"start": start,
|
|
"end": end,
|
|
"content": (
|
|
content.replace("-\n", "\n")
|
|
.replace("</u>-\n", "</u>\n")
|
|
.replace("-", " ")
|
|
.replace("%", " ")
|
|
.replace("<u> ", " <u>")
|
|
.replace(" </u>", "</u> ")
|
|
.replace("<u> </u>", "")
|
|
.replace("<u></u>", "")
|
|
.replace(" \n", "\n")
|
|
)[:-1],
|
|
}
|
|
)
|
|
# def sanitizevttwordlevel(subtitles):
|
|
# errorwords = []
|
|
# newords = {}
|
|
# for subtitle in subtitles:
|
|
# for word in subtitle["content"].split(" "):
|
|
# if ("<u>" in word):
|
|
# newword = None
|
|
# if (len(word.split("<u>")) > 1):
|
|
# newword = word.replace("<u>", " <u>")
|
|
# if (len(word.split("</u>")) > 1):
|
|
# newword = word.replace("</u>", "</u> ")
|
|
# if newword:
|
|
# original = word.split("<u>")[1].split("</u>")[0]
|
|
# if original in errorwords:
|
|
# for i in errorwords[original]:
|
|
|
|
# else:
|
|
# errorwords[orig].append(word)
|
|
|
|
# error = False
|
|
# if "<u>" in subtitle["content"]:
|
|
# if len(subtitle["content"].split("<u>")) > 1:
|
|
# if subtitle["content"].split("<u>")[0][-1] != " ":
|
|
# error = True
|
|
# if len(subtitle["content"].split("</u>")) > 1:
|
|
# if subtitle["content"].split("</u>")[1][0] != " ":
|
|
# error = True
|
|
# if error:
|
|
# word = subtitle["content"].split("<u>")[1].split("</u>")[0]
|
|
# errorwords.append(word)
|
|
# for word in subtitle["content"].split(" "):
|
|
# if word.replace("<u>")
|
|
|
|
# for subtitle in subtitles:
|
|
# for words in subtitle["content"].split(" "):
|
|
# if word in errorwords:
|
|
# subtitle["content"]
|
|
|
|
# sanitizevttwordlevel(subtitles)
|
|
return subtitles
|
|
|
|
|
|
def to_vtt(subtitles):
|
|
vtt_content = "WEBVTT\n\n\n"
|
|
for idx, subtitle in enumerate(subtitles):
|
|
content = subtitle["content"]
|
|
if not subtitle.get("split", False):
|
|
start = subtitle["start"]
|
|
end = subtitle["end"]
|
|
vtt_content += f"{start} --> {end}\n{content}\n\n\n"
|
|
else:
|
|
vtt_content += f"NOTE {content}\n\n\n"
|
|
|
|
return vtt_content.strip()
|
|
|
|
|
|
def to_stacked_vtt(subtitles):
|
|
vtt_content = "WEBVTT\n\n\n"
|
|
buffer = ""
|
|
for subtitle in subtitles:
|
|
if subtitle.get("split", False):
|
|
buffer = ""
|
|
continue
|
|
if len(buffer) != 0:
|
|
if str(subtitle["content"].strip())[-1] == ".":
|
|
buffer += "\n"
|
|
else:
|
|
buffer += " "
|
|
buffer += subtitle["content"].strip()
|
|
vtt_content += f"{subtitle['start']} --> {subtitle['end']}\n"
|
|
vtt_content += buffer
|
|
vtt_content += "\n\n\n"
|
|
return vtt_content
|
|
|
|
|
|
def script_from_word_vtt(wordvtt):
|
|
subtitles = from_vtt(wordvtt)
|
|
print(f"VTT {len(subtitles)} lines. Generating script file from VTT.")
|
|
sentences = []
|
|
EXCEPTION_FLAG, ADD_NEXT_SENTENCE = "", 0
|
|
for n, subtitle in enumerate(subtitles):
|
|
sentence = subtitle["content"].replace("<u>", "").replace("</u>", "")
|
|
if ((sentences[-1] if sentences else None) != sentence) or ADD_NEXT_SENTENCE:
|
|
sentences.append(sentence)
|
|
ADD_NEXT_SENTENCE = 0
|
|
if subtitle["content"][-4:] == "</u>":
|
|
# print(f"{len(sentences)} END {subtitle["content"]}")
|
|
ADD_NEXT_SENTENCE = 1
|
|
if n + 2 < len(subtitles):
|
|
if (
|
|
subtitles[n + 2]["content"].replace("<u>", "").replace("</u>", "")
|
|
!= sentence
|
|
):
|
|
ADD_NEXT_SENTENCE = 0
|
|
return sentences
|
|
|
|
|
|
def create_word_scenes(wordvtt, scriptraw):
|
|
subtitles = from_vtt(wordvtt)
|
|
scripts = [i for i in scriptraw.split("\n") if i]
|
|
print(f"VTT {len(subtitles)} lines, Script {len(scripts)} lines")
|
|
scenes = []
|
|
for n, script in enumerate(scripts):
|
|
if len(script.split(" ")) == 1:
|
|
continue
|
|
scenes.append({"scene": script, "timestamp": []})
|
|
|
|
scenes_cur = 0
|
|
for n, subtitle in enumerate(subtitles):
|
|
sentence = subtitle["content"].replace("<u>", "").replace("</u>", "")
|
|
if len(sentence.split(" ")) == 1:
|
|
continue
|
|
|
|
if sentence != scenes[scenes_cur].get("scene"):
|
|
if sentence == scenes[scenes_cur + 1].get("scene"):
|
|
scenes_cur += 1
|
|
else:
|
|
print(
|
|
f"Error, Mismatch in scenes\n=>\"[{scenes_cur}] {scenes[scenes_cur].get("scene")}\" or \"[{scenes_cur+1}] {scenes[scenes_cur+1].get("scene")}\" != \"{sentence}\""
|
|
)
|
|
return
|
|
|
|
current_scene = scenes[scenes_cur]
|
|
if current_scene["timestamp"]:
|
|
word_idx = current_scene["timestamp"][-1]["index"] + 1
|
|
else:
|
|
word_idx = 0
|
|
|
|
# print(scenes_cur, subtitle, word_idx, sentence)
|
|
if ("<u>" not in subtitle["content"]) and word_idx >= len(sentence.split(" ")):
|
|
pass
|
|
if ("<u>" in subtitle["content"]) and word_idx >= len(sentence.split(" ")):
|
|
print(
|
|
f"Error, index wrong. {scenes_cur}, word: {word_idx}, total words: {len(sentence.split(" "))}\n{subtitle}"
|
|
)
|
|
word_idx = 0
|
|
scenes_cur += 1
|
|
current_scene = scenes[scenes_cur]
|
|
if current_scene["timestamp"]:
|
|
word_idx = current_scene["timestamp"][-1]["index"] + 1
|
|
else:
|
|
word_idx = 0
|
|
print(f"Changed to {word_idx}, {scenes_cur}")
|
|
|
|
if "<u>" in subtitle["content"]:
|
|
# print(subtitle["content"])
|
|
word = subtitle["content"].split("<u>")[1].split("</u>")[0]
|
|
|
|
if word not in sentence.split(" "):
|
|
print(f'Error, Mismatch\n=> "{word}" not in "{sentence}"')
|
|
return
|
|
|
|
try:
|
|
assert sentence.split(" ")[word_idx] == word
|
|
except:
|
|
print(f'Error, Mismatch\n=> "{word}" != [{word_idx}] of "{sentence}"')
|
|
return
|
|
|
|
word_time = {
|
|
"start": subtitle["start"],
|
|
"end": subtitle["end"],
|
|
"index": word_idx,
|
|
"word": word,
|
|
}
|
|
current_scene["timestamp"].append(word_time)
|
|
|
|
# print(json.dumps(scenes, indent=2))
|
|
|
|
for scene in scenes:
|
|
if len(scene["scene"].split(" ")) != len(scene["timestamp"]):
|
|
print("Error, Mismatch length")
|
|
return
|
|
if "" in scene["scene"].split(" "):
|
|
print(repr(scene["scene"]))
|
|
|
|
full_script, full_scenes = [], []
|
|
for scene in scenes:
|
|
full_script += scene["scene"].split(" ")[:-1]
|
|
full_script.append(scene["scene"].split(" ")[-1] + "##")
|
|
full_scenes += scene["timestamp"]
|
|
|
|
for i, j in zip(full_script, full_scenes):
|
|
if i.replace("##", "") != j["word"]:
|
|
print("Error, Mismatch")
|
|
return
|
|
|
|
assert len(full_scenes) == len(full_script)
|
|
|
|
return full_script, full_scenes
|
|
|
|
|
|
# Detect long break or change in context, inserts section break into script.
|
|
def autobreak(lines, times):
|
|
from datetime import timedelta
|
|
|
|
def parsetime(time_str):
|
|
minutes, seconds = time_str.split(":")
|
|
seconds, milliseconds = seconds.split(".")
|
|
td = timedelta(
|
|
minutes=int(minutes), seconds=int(seconds), milliseconds=int(milliseconds)
|
|
)
|
|
return td
|
|
|
|
script = []
|
|
long_breaks = []
|
|
tmark = parsetime("0:0.0")
|
|
for i, j in zip(lines, times):
|
|
tdiff = parsetime(j["start"]) - tmark
|
|
tmark = parsetime(j["end"])
|
|
if tdiff > parsetime("0:0.0"):
|
|
long_breaks.append(tdiff)
|
|
# print()
|
|
# print(i, end=" ")
|
|
# print()
|
|
|
|
mean_break = parsetime("0:0.0")
|
|
for i in long_breaks:
|
|
mean_break += i / len(long_breaks)
|
|
print(mean_break)
|
|
|
|
script = ""
|
|
tmark = parsetime("0:0.0")
|
|
tmp = " "
|
|
|
|
continous_line = 0
|
|
for i, j in zip(lines, times):
|
|
tdiff = parsetime(j["start"]) - tmark
|
|
tmark = parsetime(j["end"])
|
|
if tdiff > mean_break and tmp[-1] != ".":
|
|
script += "\n"
|
|
|
|
if tdiff >= mean_break and tmp[-1] == ".":
|
|
script += "\n"
|
|
continous_line = 0
|
|
else:
|
|
continous_line += 1
|
|
|
|
script += i.replace("##", "")
|
|
|
|
if i[-1] == ".":
|
|
script += "\n"
|
|
elif "##" in i:
|
|
script += "\n"
|
|
else:
|
|
script += " "
|
|
tmp = i
|
|
|
|
return script
|
|
|
|
|
|
def scene_from_new_script(raw_script, full_script, full_scenes):
|
|
mod_script = raw_script.replace("\n", " \n ").split(" ")
|
|
mod_script = [i for i in mod_script if i]
|
|
n = 0
|
|
while True:
|
|
if mod_script[n] == "\n":
|
|
mod_script[n - 1] += "\n"
|
|
del mod_script[n]
|
|
n -= 1
|
|
n += 1
|
|
if n == len(mod_script):
|
|
break
|
|
# print(mod_script)
|
|
print(f"Original: {len(full_script)}, Modded: {len(mod_script)}")
|
|
allowed_list = [".", "\n", "\n\n", ",", "?", "##"]
|
|
|
|
def normalized(x):
|
|
for i in allowed_list:
|
|
x = x.replace(i, "")
|
|
return x.upper()
|
|
|
|
same = lambda a, b: normalized(a) == normalized(b)
|
|
new_script, new_timestamp, orig_index, n = [], [], 0, 0
|
|
fail = 0
|
|
while n < len(mod_script):
|
|
print(f"{repr(mod_script[n]):>20} ? {repr(full_script[orig_index])}")
|
|
word = mod_script[n]
|
|
if same(word, full_script[orig_index].replace("##", "")):
|
|
cur = full_scenes[orig_index]
|
|
new_script.append(word.replace("##", ""))
|
|
new_timestamp.append({"start": cur["start"], "end": cur["end"]})
|
|
fail = 0
|
|
else:
|
|
if fail > 10:
|
|
print("Error: Failed to match words,")
|
|
return
|
|
# print("Back")
|
|
fail += 1
|
|
n -= 1
|
|
n, orig_index = n + 1, orig_index + 1
|
|
assert len(new_script) == len(new_timestamp)
|
|
return new_script, new_timestamp
|
|
|
|
|
|
def build_new_subtitle(new_script, new_timestamp):
|
|
buffer, new_scenes, start, end = [], [], None, None
|
|
current_scene = []
|
|
# print(" ".join(new_script).split("\n"))
|
|
|
|
for i, j in zip(new_script, new_timestamp):
|
|
if "\n" in i:
|
|
buffer.append(i.replace("\n", ""))
|
|
current_scene.append(
|
|
{
|
|
"content": " ".join(buffer).replace("##", ""),
|
|
"start": start,
|
|
"end": j["end"],
|
|
}
|
|
)
|
|
buffer, start = [], None
|
|
if "\n\n" in i:
|
|
print(
|
|
f"Section break at line #{len(current_scene):<3}| \"{current_scene[-1]["content"]}\""
|
|
)
|
|
new_scenes.append(current_scene)
|
|
current_scene = []
|
|
else:
|
|
buffer.append(i)
|
|
if not start:
|
|
start = j["start"]
|
|
|
|
if start:
|
|
buffer.append(i.replace("\n", ""))
|
|
current_scene.append(
|
|
{"content": " ".join(buffer), "start": start, "end": j["end"]}
|
|
)
|
|
|
|
if current_scene != (new_scenes[-1] if new_scenes else None):
|
|
new_scenes.append(current_scene)
|
|
|
|
# print("\n\n".join(["\n".join([j["content"] for j in i]) for i in new_scenes]))
|
|
newsub = []
|
|
for n, i in enumerate(new_scenes):
|
|
newsub += i
|
|
if n < len(new_scenes) - 1:
|
|
newsub.append(
|
|
{"content": "Break", "start": None, "end": None, "split": True}
|
|
)
|
|
|
|
return newsub
|
|
|
|
|
|
def saveFile(filename, data, override=False):
|
|
if os.path.exists(filename) and not override:
|
|
print(f"File {filename} already exists.")
|
|
return -1
|
|
with open(filename, "w") as f:
|
|
f.write(data)
|
|
|
|
|
|
def openFile(filename):
|
|
with open(filename, "r") as f:
|
|
data = f.read()
|
|
if not data:
|
|
return -1
|
|
return data
|
|
|
|
|
|
def main(vttfile, scriptfile):
|
|
modfile = ".".join(scriptfile.split(".")[:-1]) + ".script"
|
|
x = create_word_scenes(openFile(vttfile), openFile(scriptfile))
|
|
if not x:
|
|
sys.exit(-1)
|
|
full_script, full_scenes = x
|
|
|
|
if not os.path.exists(modfile):
|
|
genscript = autobreak(full_script, full_scenes)
|
|
saveFile(modfile, genscript)
|
|
print(f"Saved modification file as {modfile}. Modify it and return back.")
|
|
else:
|
|
x = scene_from_new_script(openFile(modfile), full_script, full_scenes)
|
|
if not x:
|
|
sys.exit(-1)
|
|
a, b = x
|
|
|
|
final_vtt = build_new_subtitle(a, b)
|
|
jsonfile = ".".join(vttfile.split(".")[:-1]) + ".json"
|
|
saveFile(jsonfile, json.dumps(final_vtt, indent=2), True)
|
|
print(f"Saved JSON file as {jsonfile}. Fix it, and convert it to VTT.")
|
|
|
|
|
|
if __name__ == "__main__":
|
|
import sys
|
|
|
|
if len(sys.argv) not in (2, 3):
|
|
print(
|
|
f"Usage: {sys.argv[0].split("/")[-1]} [vtt file] (txt file)\n"
|
|
f" {sys.argv[0].split("/")[-1]} [JSON file]\n"
|
|
"** Only output from openai-whisper with '--word-timestamp true' is accepted.)\n"
|
|
"** You have to run this for first time, and then fix .script file, and then re-run this script.\n"
|
|
"** Adding newline/period/commas are onlt permitted. Fix else in JSON file."
|
|
)
|
|
sys.exit()
|
|
|
|
vtt = sys.argv[1]
|
|
print(f"\n[{vtt}]")
|
|
if len(sys.argv) == 3:
|
|
script = sys.argv[2]
|
|
if (not os.path.exists(vtt)) or (not os.path.exists(script)):
|
|
print(f"Input file doesnt exists.")
|
|
sys.exit(-1)
|
|
main(vtt, script)
|
|
else:
|
|
if ".json" in vtt:
|
|
final_vtt = json.loads(openFile(vtt))
|
|
orgf = ".".join(vtt.split(".")[:-1])
|
|
print(f"Saved VTT file as {orgf}.final.vtt.")
|
|
saveFile(orgf + ".final.vtt", to_vtt(final_vtt), True)
|
|
saveFile(orgf + ".stacked.vtt", to_stacked_vtt(final_vtt), True)
|
|
sys.exit(0)
|
|
if not os.path.exists(vtt):
|
|
print(f"Input file doesnt exists.")
|
|
sys.exit(-1)
|
|
script = ".".join(vtt.split(".")[:-1]) + ".txt"
|
|
saveFile(script, "\n".join(script_from_word_vtt(openFile(vtt))))
|
|
main(vtt, script)
|