mirror of
https://github.com/jxxghp/MoviePilot.git
synced 2026-09-05 15:38:19 +08:00
feat(music): 增强无标签音频识别，支持文件名/目录解析与专辑级匹配
- 新增 MusicNameParser：剥离曲序/碟号前缀、拆分歌手与曲名、解析专辑目录名（歌手/专辑/年份/音质）和 CD1 等碟片目录 - MetaInfoPath 与 MusicChain.read_path_meta 接入路径上下文，WAV 及标签不全的 FLAC/MP3 可补齐识别线索 - MusicBrainz 新增 match_music_album：按专辑名/歌手/曲名搜索候选发行版本，用曲目数、总时长、逐曲时长和曲名重合度打分对位 - MusicChain 新增 recognize_album_directory 目录级批量识别（按目录缓存），单曲识别未命中时自动兜底 - transfer 整理链路接入专辑匹配，命中后回填曲目身份用于重命名与刮削
This commit is contained in:
jxxghp
@@ -1,6 +1,7 @@
|
||||
import re
|
||||
import threading
|
||||
import time
|
||||
from difflib import SequenceMatcher
|
||||
from typing import Any, Iterable, Optional, Tuple, Union
|
||||
|
||||
from fastapi.concurrency import run_in_threadpool
|
||||
@@ -183,6 +184,254 @@ class MusicBrainzModule(_ModuleBase):
|
||||
index += 1
|
||||
return results
|
||||
|
||||
def match_music_album(
|
||||
self,
|
||||
meta: MetaMusic,
|
||||
tracks: list[MetaMusic],
|
||||
limit: int = 5,
|
||||
) -> Optional[MusicAlbumInfo]:
|
||||
"""按目录线索和曲目特征把本地音频集合对位到 MusicBrainz 发行版本。
|
||||
|
||||
适用于无标签整专目录：用专辑名、歌手搜索候选发行版本，再用曲目数、
|
||||
总时长和逐曲时长相似度打分，选出最可信的版本并返回其曲目表。
|
||||
"""
|
||||
if not tracks:
|
||||
return None
|
||||
best_album: Optional[MusicAlbumInfo] = None
|
||||
best_score = 0.0
|
||||
for release in self._search_release_candidates(meta, tracks, limit=limit):
|
||||
release_id = release.get("id")
|
||||
if not release_id:
|
||||
continue
|
||||
detail = self._request_json(
|
||||
f"/release/{release_id}",
|
||||
params={"inc": "recordings+media+artist-credits", "fmt": "json"},
|
||||
)
|
||||
if not detail:
|
||||
continue
|
||||
summary = self._release_track_summary(detail)
|
||||
score = self._score_release(meta, tracks, detail, summary)
|
||||
if score > best_score:
|
||||
best_score = score
|
||||
best_album = self._release_to_album(detail)
|
||||
# 得分低于阈值时宁可不匹配，避免把曲目写到错误的专辑上
|
||||
if best_score < self._album_match_threshold:
|
||||
return None
|
||||
return best_album
|
||||
|
||||
_album_match_threshold = 60.0
|
||||
|
||||
def _search_release_candidates(
|
||||
self,
|
||||
meta: MetaMusic,
|
||||
tracks: list[MetaMusic],
|
||||
limit: int,
|
||||
) -> list[dict[str, Any]]:
|
||||
"""按专辑名和曲名线索搜索候选发行版本，多个查询按命中顺序去重。"""
|
||||
releases: list[dict[str, Any]] = []
|
||||
seen: set[str] = set()
|
||||
for query in self._release_queries(meta, tracks):
|
||||
payload = self._request_json(
|
||||
"/release",
|
||||
params={"query": query, "limit": max(1, min(limit, 25)), "fmt": "json"},
|
||||
)
|
||||
for item in (payload or {}).get("releases") or []:
|
||||
release_id = item.get("id")
|
||||
if release_id and release_id not in seen:
|
||||
seen.add(release_id)
|
||||
releases.append(item)
|
||||
if len(releases) >= limit:
|
||||
break
|
||||
return releases[:limit]
|
||||
|
||||
@classmethod
|
||||
def _release_queries(cls, meta: MetaMusic, tracks: list[MetaMusic]) -> list[str]:
|
||||
"""构造专辑搜索表达式：优先专辑名+歌手，无专辑线索时用曲名兜底。"""
|
||||
queries: list[str] = []
|
||||
album_title = meta.album or meta.title
|
||||
artist = meta.artists[0] if meta.artists else meta.album_artist
|
||||
if album_title:
|
||||
if artist:
|
||||
queries.append(
|
||||
f'release:"{cls._escape_query(album_title)}" AND artist:"{cls._escape_query(artist)}"'
|
||||
)
|
||||
queries.append(f'release:"{cls._escape_query(album_title)}"')
|
||||
# 目录名无意义时（如 Various Artists 合集），用代表性曲名反查所属发行版本
|
||||
titles = cls._unique_texts(
|
||||
[track.title for track in tracks if track.title and not track.title.strip().isdigit()]
|
||||
)[:3]
|
||||
if titles:
|
||||
recording_clause = " OR ".join(
|
||||
f'recording:"{cls._escape_query(title)}"' for title in titles
|
||||
)
|
||||
query = f"({recording_clause})"
|
||||
if artist:
|
||||
query += f' AND artist:"{cls._escape_query(artist)}"'
|
||||
queries.append(query)
|
||||
return queries
|
||||
|
||||
@classmethod
|
||||
def _release_track_summary(cls, detail: dict[str, Any]) -> list[dict[str, Any]]:
|
||||
"""提取发行版本的曲目概要（碟号、曲序、时长、标题）供打分使用。"""
|
||||
summary: list[dict[str, Any]] = []
|
||||
for medium in detail.get("media") or []:
|
||||
disc = cls._optional_int(medium.get("position")) or 1
|
||||
for track in medium.get("tracks") or []:
|
||||
recording = track.get("recording") or {}
|
||||
summary.append({
|
||||
"disc": disc,
|
||||
"position": cls._optional_int(track.get("position")),
|
||||
"length": cls._duration_seconds(
|
||||
track.get("length") or recording.get("length")
|
||||
),
|
||||
"title": track.get("title") or recording.get("title"),
|
||||
})
|
||||
return summary
|
||||
|
||||
@classmethod
|
||||
def _score_release(
|
||||
cls,
|
||||
meta: MetaMusic,
|
||||
tracks: list[MetaMusic],
|
||||
detail: dict[str, Any],
|
||||
summary: list[dict[str, Any]],
|
||||
) -> float:
|
||||
"""给候选发行版本打分（0-100），综合标题、歌手、曲目数和时长相似度。"""
|
||||
local_count = len(tracks)
|
||||
release_count = len(summary)
|
||||
if not release_count:
|
||||
return 0.0
|
||||
# 曲目数差异过大直接排除，避免单曲误命中整专或反之
|
||||
diff = abs(local_count - release_count)
|
||||
if diff > max(4, int(local_count * 0.5)):
|
||||
return 0.0
|
||||
# 本地文件比发行版本多出的曲目无法被覆盖，超出容忍范围视为错误候选
|
||||
if local_count > release_count and diff > max(1, int(release_count * 0.25)):
|
||||
return 0.0
|
||||
score = 0.0
|
||||
# 标题相似度：专辑目录名或文件标签中的专辑名/曲名
|
||||
title_hints = cls._unique_texts([meta.album, meta.title])
|
||||
title_sim = max(
|
||||
(cls._text_similarity(hint, detail.get("title")) for hint in title_hints),
|
||||
default=0.0,
|
||||
)
|
||||
artist_names = cls._artist_credits(detail.get("artist-credit"))[0]
|
||||
if meta.artists and artist_names:
|
||||
artist_sim = max(
|
||||
cls._text_similarity(meta.artists[0], name) for name in artist_names
|
||||
)
|
||||
score += 40 * title_sim + 15 * artist_sim
|
||||
else:
|
||||
# 缺少歌手线索时把权重让给标题
|
||||
score += 50 * title_sim
|
||||
# 曲目数：完全一致是最强信号
|
||||
if diff == 0:
|
||||
score += 15
|
||||
elif diff == 1:
|
||||
score += 8
|
||||
elif diff <= max(2, int(local_count * 0.15)):
|
||||
score += 2
|
||||
# 曲名重合度：部分曲目目录（只下载了整专的一部分）依靠曲名对位确认
|
||||
release_titles = {cls._match_text(item["title"]) for item in summary}
|
||||
named_tracks = [track for track in tracks if track.title and not track.title.strip().isdigit()]
|
||||
if named_tracks and release_titles:
|
||||
overlap = sum(
|
||||
1 for track in named_tracks if cls._match_text(track.title) in release_titles
|
||||
)
|
||||
score += 15 * overlap / len(named_tracks)
|
||||
# 总时长：无损整专 rip 的总时长与 MusicBrainz 记录高度接近
|
||||
local_total = sum(track.duration or 0 for track in tracks)
|
||||
release_total = sum(item["length"] or 0 for item in summary)
|
||||
local_durations = [track.duration for track in tracks if track.duration]
|
||||
if local_durations and release_total:
|
||||
delta = abs(local_total - release_total) / max(local_total, release_total)
|
||||
if delta <= 0.02:
|
||||
score += 15
|
||||
elif delta <= 0.05:
|
||||
score += 10
|
||||
elif delta <= 0.10:
|
||||
score += 5
|
||||
# 逐曲时长对位：曲目数一致时逐首比较
|
||||
if diff == 0 and len(local_durations) == local_count:
|
||||
similarities = []
|
||||
for track, item in zip(
|
||||
sorted(tracks, key=lambda item: (item.disc_number or 1, item.track_number or 0)),
|
||||
summary,
|
||||
):
|
||||
if track.duration and item["length"]:
|
||||
similarities.append(cls._duration_similarity(track.duration, item["length"]))
|
||||
if similarities:
|
||||
score += 15 * sum(similarities) / len(similarities)
|
||||
return score
|
||||
|
||||
@staticmethod
|
||||
def _duration_similarity(left: int, right: int) -> float:
|
||||
"""比较两个时长的接近程度，完全一致为 1，差异越大越接近 0。"""
|
||||
if not left or not right:
|
||||
return 0.0
|
||||
return max(0.0, 1 - abs(left - right) / max(left, right))
|
||||
|
||||
@classmethod
|
||||
def _text_similarity(cls, left: Optional[str], right: Optional[str]) -> float:
|
||||
"""忽略大小写和标点后比较两段音乐文本的相似度。"""
|
||||
normalized_left = cls._match_text(left)
|
||||
normalized_right = cls._match_text(right)
|
||||
if not normalized_left or not normalized_right:
|
||||
return 0.0
|
||||
return SequenceMatcher(None, normalized_left, normalized_right).ratio()
|
||||
|
||||
@staticmethod
|
||||
def _match_text(value: Optional[str]) -> str:
|
||||
"""移除大小写、空白和标点差异，生成相似度比较使用的紧凑文本。"""
|
||||
return re.sub(r"[\W_]+", "", str(value or "").casefold(), flags=re.UNICODE)
|
||||
|
||||
@classmethod
|
||||
def _unique_texts(cls, values: Iterable[Optional[str]]) -> list[str]:
|
||||
"""按规范化文本去重并保留原始顺序。"""
|
||||
results: list[str] = []
|
||||
seen: set[str] = set()
|
||||
for value in values:
|
||||
normalized = cls._normalize_text(value)
|
||||
identity = normalized.casefold()
|
||||
if not normalized or identity in seen:
|
||||
continue
|
||||
seen.add(identity)
|
||||
results.append(normalized)
|
||||
return results
|
||||
|
||||
@classmethod
|
||||
def _release_to_album(cls, detail: dict[str, Any]) -> Optional[MusicAlbumInfo]:
|
||||
"""将 MusicBrainz Release 详情转换为带曲目表的标准化专辑信息。"""
|
||||
release_id = detail.get("id")
|
||||
title = detail.get("title")
|
||||
if not release_id or not title:
|
||||
return None
|
||||
release_group = detail.get("release-group") or {}
|
||||
group_id = release_group.get("id")
|
||||
artists, artist_ids = cls._artist_credits(detail.get("artist-credit"))
|
||||
album = MusicAlbumInfo(
|
||||
source=cls._source,
|
||||
# 优先使用 Release Group ID，与专辑详情和封面入口保持一致
|
||||
media_id=str(group_id or release_id),
|
||||
title=str(title),
|
||||
artists=artists,
|
||||
artist_ids=artist_ids,
|
||||
album_type=release_group.get("primary-type"),
|
||||
secondary_types=[str(item) for item in release_group.get("secondary-types") or []],
|
||||
release_date=detail.get("date") or None,
|
||||
cover_url=cls._build_cover_url(group_id),
|
||||
genres=cls._names_of(detail.get("genres")),
|
||||
detail_link=f"https://musicbrainz.org/release/{release_id}",
|
||||
raw_data={"release_id": str(release_id)},
|
||||
)
|
||||
album.tracks = [
|
||||
info
|
||||
for medium in detail.get("media") or []
|
||||
for track in medium.get("tracks") or []
|
||||
if (info := cls._track_to_info(album, medium, track))
|
||||
]
|
||||
return album
|
||||
|
||||
def recognize_media(
|
||||
self,
|
||||
meta: MetaBase = None,
|
||||
|
||||
Reference in New Issue
Block a user