Shenrui-Ma/video-regen-recipes / templates/singing-cover/character-rvc
character-rvc-training-cover
Verified为用户指定角色发现与核验语音来源,整理授权录音和分组留出集,在固定 RVC v2/48k/F0 环境训练与选择模型、校验 added 索引,再分声部转换歌曲并混音交付;莫斯提马中文若舞仅为历史示例。
| File | Kind | Size |
|---|---|---|
| LICENSE | other | 1.1 KB |
| LICENSES.md | markdown | 1.8 KB |
| README.md | markdown | 5.2 KB |
| SKILL.md | skill md | 5.5 KB |
| examples/index-inputs.spec.json | config | 471 B |
| model-catalog/dependencies.json | config | 3.0 KB |
| profile.json | config | 3.3 KB |
| references/deployment.md | markdown | 6.8 KB |
| references/discovery.md | markdown | 9.6 KB |
| references/processing.md | markdown | 10.5 KB |
| references/repair.md | markdown | 7.8 KB |
| references/training.md | markdown | 12.0 KB |
| references/validation.md | markdown | 6.8 KB |
| scripts/build_filelist.py | script | 5.5 KB |
| scripts/discover.py | script | 13.1 KB |
| scripts/run_guard.py | script | 4.9 KB |
| sources.md | markdown | 4.7 KB |
| tests/test_discover.py | script | 11.3 KB |
| tests/test_helpers.py | script | 6.9 KB |
scripts/build_filelist.py
View on GitHub#!/usr/bin/env python3
"""Build a v2/48k/F0/speaker-0 filelist from a four-modality intersection."""
import argparse
import hashlib
import json
from pathlib import Path
import random
import sys
from run_guard import require_regular, resource_path, sha256, write_new_json
MODALITIES = {
"wav": ("0_gt_wavs", ".wav"),
"feature768": ("3_feature768", ".npy"),
"f0": ("2a_f0", ".wav.npy"),
"f0nsf": ("2b-f0nsf", ".wav.npy"),
}
def safe_text(value):
if any(char in str(value) for char in ("|", "\n", "\r")):
raise ValueError("Filelist paths cannot contain pipes or newlines")
def collect(experiment):
modalities = {}
for key, (directory, suffix) in MODALITIES.items():
folder = experiment / directory
if not folder.is_dir() or folder.is_symlink():
raise ValueError("Missing or linked modality directory: " + directory)
files = {}
for path in sorted(folder.iterdir()):
if not path.name.endswith(suffix):
continue
safe_text(path)
require_regular(path)
name = path.name[:-len(suffix)]
if not name:
raise ValueError("Empty sample ID")
files[name] = path
modalities[key] = files
return modalities
def build(experiment, mute_root, allow_incomplete=False):
modalities = collect(experiment)
sets = [set(files) for files in modalities.values()]
union = set.union(*sets)
intersection = set.intersection(*sets)
if not intersection:
raise ValueError("No complete character samples")
excluded = {name: [key for key, files in modalities.items() if name not in files]
for name in sorted(union - intersection)}
if excluded and not allow_incomplete:
raise ValueError("Incomplete modalities; review missing IDs before --allow-incomplete: "
+ json.dumps(excluded, sort_keys=True))
mute = [mute_root / "0_gt_wavs/mute48k.wav",
mute_root / "3_feature768/mute.npy",
mute_root / "2a_f0/mute.wav.npy",
mute_root / "2b-f0nsf/mute.wav.npy"]
sample_paths = [[modalities[key][name] for key in MODALITIES] for name in sorted(intersection)]
# A mute fixture is never counted as a character sample, including hard links.
for path in mute:
safe_text(path)
resource_path(mute_root, str(path))
require_regular(path)
if any(path.samefile(candidate) for row in sample_paths for candidate in row):
raise ValueError("Mute and character files overlap")
rows = ["|".join(map(str, paths)) + "|0" for paths in sample_paths]
random.Random(20260731).shuffle(rows)
mute_row = "|".join(map(str, mute)) + "|0"
text = "\n".join(rows + [mute_row, mute_row]) + "\n"
hashes = {name: {key: sha256(modalities[key][name]) for key in MODALITIES}
for name in sorted(intersection)}
report = {
"schema_version": 1, "profile": "v2-48k-f0-speaker0",
"modality_counts": {key: len(files) for key, files in modalities.items()},
"character_rows": len(rows), "mute_rows": 2, "total_rows": len(rows) + 2,
"excluded_missing_modalities": excluded, "sample_sha256": hashes,
"mute_sha256": {key: sha256(path) for key, path in zip(MODALITIES, mute)},
"filelist_seed": 20260731, "training_seed": "not-controlled-by-this-helper",
"filelist_sha256": hashlib.sha256(text.encode("utf-8")).hexdigest(),
"not_checked": ["audio decoding", "NPY shape, finiteness and frame alignment",
"speaker identity", "holdout isolation", "authorization", "index internals"],
}
return text, report
def main(argv=None):
parser = argparse.ArgumentParser(description=__doc__)
parser.add_argument("--experiment", type=Path, required=True)
parser.add_argument("--mute-root", type=Path, required=True)
parser.add_argument("--output", type=Path, required=True)
parser.add_argument("--report", type=Path, required=True)
parser.add_argument("--allow-incomplete", action="store_true",
help="Explicitly accept the reported incomplete-sample exclusions")
args = parser.parse_args(argv)
try:
experiment = resource_path(Path.cwd(), str(args.experiment))
mute_root = resource_path(Path.cwd(), str(args.mute_root))
outputs = [args.output.resolve(), args.report.resolve()]
if outputs[0] == outputs[1] or any(path.exists() for path in outputs):
raise ValueError("Output and report must be distinct new files")
for path in outputs:
if path.is_relative_to(mute_root) or any(
path.is_relative_to(experiment / directory) for directory, _ in MODALITIES.values()
):
raise ValueError("Do not write outputs into modality or mute directories")
text, report = build(experiment, mute_root, args.allow_incomplete)
# Exclusive creation prevents clobbering. A partial pair is not a valid receipt.
with args.output.open("x", encoding="utf-8", newline="\n") as stream:
stream.write(text)
write_new_json(args.report, report)
print(json.dumps({"status": "built", "character_rows": report["character_rows"],
"mute_rows": 2, "excluded": len(report["excluded_missing_modalities"])}))
return 0
except (OSError, ValueError, TypeError) as error:
print(json.dumps({"status": "error", "error": str(error)}), file=sys.stderr)
return 2
if __name__ == "__main__":
raise SystemExit(main())
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack." | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read. | Stars: 196k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report. | Stars: 196k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt. | Stars: 196k | Updated | Claude CodeCodex | Verified | Install |