Voice mother with Carol and synchronize all call events

This commit is contained in:
Codex
2026-10-30 00:43:57 +02:00
parent 23d4e49468
commit eaefde2e07
124 changed files with 7903 additions and 2241 deletions
+15 -2
View File
@@ -1,6 +1,6 @@
"""Split a voice batch against word timestamps; reject missing script words.
Usage: python split_voice_batch.py AUDIO WORDS_JSON LINES_JSON OUTPUT_DIRECTORY
Usage: python split_voice_batch.py AUDIO WORDS_JSON LINES_JSON OUTPUT_DIRECTORY [SCORE_JSON]
Requires soundfile (the local build helper directory is also supported).
"""
from pathlib import Path
@@ -26,11 +26,24 @@ def main():
data = json.loads(words_path.read_text(encoding="utf-8"))
words = data["words"] if isinstance(data, dict) else data
lines = json.loads(lines_path.read_text(encoding="utf-8"))
pronunciations = {}
if len(sys.argv) > 5:
score = json.loads(Path(sys.argv[5]).read_text(encoding="utf-8"))
pronunciations = score["pronunciation"]
spoken, timing = [], []
in_tag = False
for word in words:
if word.get("type", "word") != "word":
continue
for token in tokens(word["text"]):
text = word["text"]
# The provider may retain direction tags and IPA in its timed output.
# Tags are metadata, not dialogue; map IPA back to the caption word.
if text.startswith("[") or in_tag:
in_tag = "]" not in text
continue
for spelling, ipa in pronunciations.items():
text = text.replace(ipa, spelling)
for token in tokens(text):
spoken.append(token)
timing.append((float(word["start"]), float(word["end"])))
reference, spans = [], []