Add no-filler tags, media CLI, and ask-before-edit iron law.
selftest / selftest (push) Canceled after 0s

Default solo-lead tags and report-only filler_ratio; slim Agent docs with media CLI; require consent, git backup or plugin path, and explicit commit/push before code changes.

Co-authored-by: Cursor <cursoragent@cursor.com>
This commit is contained in:
swl
2026-10-10 02:35:42 +08:00
co-authored by Cursor
parent dbfb9b52f9
commit 0ab8d8ad4d
13 changed files with 464 additions and 72 deletions
+4 -1
View File
@@ -3,7 +3,7 @@
命令:
doctor / styles / target / check / gen / runs / show / catalog /
prune / qc / selftest
prune / media / qc / selftest
设计约束:所有业务参数来自配置/命令行,本文件不写死任何模型或采样常量。
实现见 common / assemble / commands / selftest / qc。
@@ -146,6 +146,9 @@ def build_parser():
help="一并删除 tmp/_migrated 等非日期目录")
pr.set_defaults(func=cmds.cmd_prune)
from .media import add_media
add_media(sub)
from .qc import add_qc
add_qc(sub)
return p
+107
View File
@@ -0,0 +1,107 @@
# -*- coding: utf-8 -*-
"""songforge media:info / convert / clip(经 qc_python 子进程)。"""
import json
import os
import subprocess
from ..config import loader
from ..core import paths as pmod
from .qc import find_qc_python
ROOT = pmod.PKG_ROOT
WORKER = os.path.join(ROOT, "tools", "media_worker.py")
def _run(job, timeout=600):
cfg = loader.load({})
py = find_qc_python(cfg)
if not py:
print("[错误] 找不到 qc_python。请在 config/local.toml 配置 paths.qc_python")
return 2, None
env = dict(os.environ)
cro = pmod.comfy_root(cfg=cfg)
env["SONGFORGE_COMFY_ROOT"] = cro
env["PYTHONIOENCODING"] = "utf-8"
cmd = [py, WORKER, json.dumps(job, ensure_ascii=False)]
try:
r = subprocess.run(cmd, capture_output=True, env=env, timeout=timeout)
except subprocess.TimeoutExpired:
print("[错误] media 超时")
return 2, None
out = r.stdout.decode("utf-8", "replace")
if "###JSON###" not in out:
print(out[-1500:])
print(r.stderr.decode("utf-8", "replace")[-1500:])
print("[错误] media_worker 无 JSON 输出")
return 2, None
blob = json.loads(out.split("###JSON###", 1)[1].strip())
if blob.get("error"):
print("[错误] %s: %s" % (blob.get("type"), blob["error"]))
return 1, blob
return 0, blob
def cmd_media_info(args):
code, blob = _run(dict(op="info", file=os.path.abspath(args.file)))
if blob and not blob.get("error"):
print(json.dumps(blob, ensure_ascii=False, indent=1))
return code
def cmd_media_convert(args):
out = args.out
if not out:
base, _ = os.path.splitext(args.file)
out = base + "." + args.to.lstrip(".")
code, blob = _run(dict(
op="convert", file=os.path.abspath(args.file),
to=args.to, out=os.path.abspath(out)))
if blob and not blob.get("error"):
print(json.dumps(blob, ensure_ascii=False, indent=1))
return code
def cmd_media_clip(args):
out = args.out
if not out:
base, ext = os.path.splitext(args.file)
out = "%s_clip%.1f%s" % (base, args.start, ext or ".wav")
job = dict(op="clip", file=os.path.abspath(args.file),
start=args.start, out=os.path.abspath(out))
if args.dur is not None:
job["dur"] = args.dur
if args.end is not None:
job["end"] = args.end
code, blob = _run(job)
if blob and not blob.get("error"):
print(json.dumps(blob, ensure_ascii=False, indent=1))
return code
def add_media(sub):
m = sub.add_parser("media", help="音频窄接口:info / convert / clip")
ms = m.add_subparsers(dest="media_cmd")
i = ms.add_parser("info", help="时长/采样率/声道/粗 BPM")
i.add_argument("--file", required=True)
i.set_defaults(func=cmd_media_info)
c = ms.add_parser("convert", help="转 wav/flac/mp3")
c.add_argument("--file", required=True)
c.add_argument("--to", required=True, choices=["wav", "flac", "mp3"])
c.add_argument("--out")
c.set_defaults(func=cmd_media_convert)
k = ms.add_parser("clip", help="按秒截取")
k.add_argument("--file", required=True)
k.add_argument("--start", type=float, default=0.0)
k.add_argument("--dur", type=float, help="时长(秒)")
k.add_argument("--end", type=float, help="结束时刻(秒),与 --dur 二选一")
k.add_argument("--out")
k.set_defaults(func=cmd_media_clip)
def _need_sub(args):
print("用法: songforge media {info,convert,clip} ...")
return 2
m.set_defaults(func=_need_sub)
return m
+7 -5
View File
@@ -333,17 +333,19 @@ def _render(rep, args, rundir=None, cfg=None):
% (lk["t0"], hit, lk.get("text", "")))
print("\n== 声线 / 歌词 ==")
print("%-22s %-14s %8s %7s %7s %6s %6s %7s"
% ("file", "最近邻老歌", "NN相似", "自相似", "cov", "f0中位",
"f0漂移", "t漂移"))
print("%-22s %-14s %8s %7s %7s %6s %6s %6s %7s"
% ("file", "最近邻老歌", "NN相似", "自相似", "cov", "衬词%",
"f0中位", "f0漂移", "t漂移"))
for r in rows:
if "error" in r:
continue
print("%-22s %-14s %8.3f %7.3f %6.1f%% %6.1f %5.1fst %6.1f%%"
print("%-22s %-14s %8.3f %7.3f %6.1f%% %5.0f%% %6.1f %5.1fst %6.1f%%"
% (r["file"][:22], (r.get("voice_nn") or "-")[:14],
r.get("voice_nn_sim", 0), r.get("voice_self", 0),
r.get("coverage", 0) * 100, r.get("f0_med", 0),
r.get("coverage", 0) * 100, r.get("filler_ratio", 0) * 100,
r.get("f0_med", 0),
r.get("f0_drift_st", 0), r.get("tempo_drift", 0) * 100))
print(" (衬词%% = ASR 语气/哼唱占比,弱参考,不单独否决)")
print("\n== 结论 ==")
npass = 0
+20 -8
View File
@@ -124,8 +124,14 @@ def _t_model_locked():
def _t_ironlaw():
tags = ironlaw.build_tags("Chinese pop")
for kw in ("swlvoice", "8 bar instrumental intro hook", "no opening vocals",
"no pure vocal intro", "no humming"):
"no pure vocal intro", "no humming",
"solo lead vocal only", "no backing vocals",
"sing only the exact words"):
assert kw in tags, "tags 缺少 %s" % kw
assert "call and response" not in tags.lower() or \
"no call and response vocals" in tags.lower(), \
"MEMO 仍鼓励 call and response 且未禁叠唱"
assert "single lead vocal hook" in tags, "缺少单人主唱 hook 描述"
lyrics = ironlaw.build_lyrics(
[("[verse]", "第一句歌词\n第二句歌词")], "kick and snare", bars=8)
assert lyrics.lstrip().startswith("[intro]"), "未强制器乐前奏"
@@ -137,7 +143,12 @@ def _t_ironlaw():
# 中文结构标记必须被拦
marked = lyrics.replace("[verse]", "[主歌]")
assert ironlaw.check_rules(marked, tags), "中文结构标记未被拦截"
return [("铁律注入", "8 小节器乐前奏 + 反人声条款齐备,缺前奏/中文标记均被拦")]
# lean 也必须带衬音硬禁
lean = ironlaw.build_tags("Chinese pop", lean=True)
for kw in ("solo lead vocal only", "no backing vocals"):
assert kw in lean, "lean tags 缺少衬音禁令 %s" % kw
return [("铁律注入",
"8 小节前奏 + 无衬音/无和声叠唱 + lean 齐备,缺前奏/中文标记均被拦")]
def _t_sparse_profile():
@@ -338,15 +349,16 @@ def _t_chill_delay():
# 但也不能把人声彻底禁掉,否则就是纯器乐、不是吟唱。
# 注意只匹配纯器乐专属的反人声条款:
# - 不能用裸 "no singing":"no singing in any breakdown" 含这个子串,
# 那是吟唱要的人声约束。
# - 不能用 "instrumental only":它来自 "intro is instrumental only",
# 那是前奏必须保留的约束。
for kw in ("no vocals anywhere", "no voice at all", "no lead vocal",
"no backing vocal", "this is a backing track",
# - 不能用裸 "no singing":"no singing in any breakdown" 含这个子串。
# - 不能用 "no backing vocal"(无 s):那是 INST_NEG;默认衬音禁令是
# "no backing vocals"(有 s)+ "solo lead vocal only",chill 必须保留。
for kw in ("no vocals anywhere", "no voice at all", "no lead vocal,",
"no backing vocal,", "this is a backing track",
"drums bass synths chords and lead melodies only"):
assert kw not in tags.lower(), \
"chill tags 含纯器乐条款 %s,吟唱被人声禁令吞掉" % kw
assert "no backing vocals" in tags.lower(), \
"chill 缺少默认无和声叠唱禁令"
return [("吟唱标签梯度",
"时机类措辞全部禁掉,密度触发词保留,"