小鱼开发
e15bdaf996
- 修复 duration 解析 bug (parseInt→parseFloat)，解决素材'换一个'候选池过小 - 素材匹配策略：候选池=满足时长+最近5个，严格模式排除已用素材 - Step2 下一步按钮绑定 dubbingAudioUrl 生成状态 - 修复 VoiceDubbing 生成后未同步 projectStore - 修复 _meta.json JSON 格式错误导致分类列表空白 - Step3/Step4 视频预览区添加标题 - 压制字幕按钮固定在底部 - 选项卡按钮高度微调
116 lines
3.7 KiB
Python
116 lines
3.7 KiB
Python
"""
|
|
空镜素材服务
|
|
============
|
|
|
|
从本地 JSON 配置文件加载素材库，提供匹配逻辑。
|
|
duration 从文件名 `_{N}s_` 中自动解析。
|
|
"""
|
|
|
|
import json
|
|
import random
|
|
import re
|
|
from pathlib import Path
|
|
|
|
# 正则：从文件名中提取时长，如 plumbing_10s_a23f8fcb.mp4 → 10
|
|
_DURATION_RE = re.compile(r"_(\d+)s_")
|
|
|
|
# 配置缓存（启动时加载，运行时只读）
|
|
_keywords: dict[str, str] = {}
|
|
_materials: dict[str, list[dict]] = {}
|
|
|
|
|
|
def _get_config_path() -> Path:
|
|
"""获取配置文件绝对路径"""
|
|
# 配置文件位于项目根目录的 config/ 下
|
|
return Path(__file__).resolve().parent.parent.parent / "config" / "materials.json"
|
|
|
|
|
|
def _parse_duration(url: str) -> float:
|
|
"""从 URL 文件名中解析时长（秒）"""
|
|
filename = url.split("/")[-1]
|
|
match = _DURATION_RE.search(filename)
|
|
if not match:
|
|
raise ValueError(f"无法从文件名解析时长: {filename}")
|
|
return float(match.group(1))
|
|
|
|
|
|
def load_config() -> None:
|
|
"""加载素材配置到内存缓存"""
|
|
global _keywords, _materials
|
|
|
|
config_path = _get_config_path()
|
|
if not config_path.exists():
|
|
raise FileNotFoundError(f"素材配置文件不存在: {config_path}")
|
|
|
|
with open(config_path, "r", encoding="utf-8") as f:
|
|
data = json.load(f)
|
|
|
|
_keywords = data.get("keywords", {})
|
|
raw_materials = data.get("materials", {})
|
|
|
|
# 解析每个素材的 duration
|
|
_materials = {}
|
|
for slug, entries in raw_materials.items():
|
|
parsed = []
|
|
for entry in entries:
|
|
url = entry["url"]
|
|
duration = _parse_duration(url)
|
|
parsed.append({"url": url, "duration": duration})
|
|
_materials[slug] = parsed
|
|
|
|
|
|
def match_material(scene: str, required_duration: float, exclude_urls: list[str] | None = None, strict: bool = False) -> dict | None:
|
|
"""
|
|
根据场景描述和所需时长匹配空镜素材
|
|
|
|
策略：
|
|
1. 收集所有满足时长要求（duration >= required_duration）的素材
|
|
2. 收集全局差值最近的 5 个素材
|
|
3. 合并去重后从候选池中随机选取，优先排除已使用的
|
|
|
|
Args:
|
|
scene: 分镜场景描述
|
|
required_duration: 所需时长（秒）
|
|
exclude_urls: 已使用的素材 URL 列表（避免重复）
|
|
|
|
Returns:
|
|
匹配到的素材 {url, duration}，无匹配返回 None
|
|
"""
|
|
exclude_urls = exclude_urls or []
|
|
for keyword, slug in _keywords.items():
|
|
if keyword in scene:
|
|
all_materials = _materials.get(slug, [])
|
|
if not all_materials:
|
|
return None
|
|
|
|
# 1. 满足时长要求的素材
|
|
matching = [m for m in all_materials if m["duration"] >= required_duration]
|
|
|
|
# 2. 差值最近的 5 个素材（全局）
|
|
sorted_by_diff = sorted(all_materials, key=lambda m: abs(m["duration"] - required_duration))
|
|
closest_5 = sorted_by_diff[:5]
|
|
|
|
# 3. 合并候选池并去重（matching 在前，优先保留满足时长的）
|
|
candidate_pool = []
|
|
seen = set()
|
|
for m in matching + closest_5:
|
|
if m["url"] not in seen:
|
|
candidate_pool.append(m)
|
|
seen.add(m["url"])
|
|
|
|
# 4. 排除已使用的，从中随机选
|
|
unused = [m for m in candidate_pool if m["url"] not in exclude_urls]
|
|
if unused:
|
|
return random.choice(unused)
|
|
|
|
# 5. 严格模式下不允许返回已排除的素材
|
|
if strict:
|
|
return None
|
|
|
|
# 6. 非严格模式：全部用完则允许重复
|
|
if candidate_pool:
|
|
return random.choice(candidate_pool)
|
|
|
|
return None
|
|
return None
|