Files
Diabolo/DoOtboy/tools/install_continuity_voices.py
T

82 lines
4.2 KiB
Python

"""Install word-verified continuity revisions, preserving all other voice takes."""
from pathlib import Path
import hashlib
import json
import math
import runpy
import shutil
import sys
import argparse
ROOT = Path(__file__).resolve().parents[1]
GAME = ROOT / "do-otboy"
sys.path.insert(0, str(ROOT / "builds/python_audio"))
import soundfile as sf
def read(path):
return json.loads(path.read_text(encoding="utf-8"))
def write(path, value):
path.write_text(json.dumps(value, ensure_ascii=False, indent=2) + "\n", encoding="utf-8")
def main():
parser = argparse.ArgumentParser()
parser.add_argument("--batch", default="continuity")
parser.add_argument("--count", type=int, default=18)
parser.add_argument("--version", default="0.2.11")
args = parser.parse_args()
assert args.batch and all(c.isalnum() or c in "_-" for c in args.batch)
work = ROOT / "builds" / args.batch
story = read(GAME / "data/phone_story.json")
metadata = read(GAME / "data/voice_manifest.json")
assembly = read(work / "assembly.json")
prepared = []
for character in ("katya", "mother"):
for segment in read(work / (character + "_split/alignment.json")):
key = segment["id"]
assert segment["text"] == story["lines"][key]["text"], key
assert segment["reference_words"] == segment["matched_words"], key
path = work / (character + "_split") / segment["file"]
info = sf.info(path)
assert info.channels == 1 and info.subtype == "PCM_16" and info.duration > .2
prepared.append((character, segment, path))
assert len(prepared) == args.count
for _, segment, path in prepared:
shutil.copyfile(path, GAME / "assets/voice" / segment["file"])
runpy.run_path(str(ROOT / "tools/write_phone_story.py"))
runpy.run_path(str(ROOT / "tools/write_voice_direction.py"))
score = {entry["id"]: entry for entry in read(ROOT / "voiceover/performance.json")["lines"]}
alignment = read(ROOT / "voiceover/alignment.json")
changed = {segment["id"] for _, segment, _ in prepared}
for batch in alignment:
alignment[batch] = [entry for entry in alignment[batch] if entry["id"] not in changed]
levels = read(GAME / "data/voice_levels.json")
for character, segment, path in prepared:
key = segment["id"]
original = metadata.get(key) or next(e for e in metadata.values() if e["speaker"] == score[key]["speaker"])
entry = {field: original[field] for field in ("provider", "voice", "voice_id", "voice_name", "speaker")}
entry.update(assembly[character], flow_id=assembly["flow_id"], id=key,
text=score[key]["text"], direction=score[key]["direction"],
tts_text=score[key]["tts_text"], pronunciation=score[key]["pronunciation"],
file="assets/voice/" + path.name, source_start=segment["start"],
source_end=segment["end"], wav_duration=segment["duration"],
sha256=hashlib.sha256(path.read_bytes()).hexdigest())
metadata[key] = entry
samples, _ = sf.read(path)
levels[key] = round(20 * math.log10(max(1e-8, float((samples ** 2).mean()) ** .5)), 3)
alignment.setdefault(args.batch + "_" + character, []).append(segment)
assert all(metadata[key]["text"] == line["text"] for key, line in story["lines"].items())
write(GAME / "data/voice_manifest.json", metadata)
write(GAME / "data/voice_levels.json", levels)
write(ROOT / "voiceover/alignment.json", alignment)
for character, speaker in (("carol", "Мама"), ("jessica", "Катя")):
entries = [dict(e, file="../../do-otboy/" + e["file"]) for e in metadata.values() if e["speaker"] == speaker]
write(ROOT / "voiceover" / character / "manifest.json", dict(
status="complete", version=args.version, voice_id=entries[0]["voice_id"],
voice_name=entries[0]["voice_name"], ready=entries, pending=[]))
write(ROOT / "voiceover" / character / "alignment.json", [e for batch in alignment.values() for e in batch if metadata[e["id"]]["speaker"] == speaker])
print(f"Installed {len(prepared)} verified lines; preserved {len(metadata)-len(prepared)} other recordings")
if __name__ == "__main__":
main()