yuanflow-cli 0.1.10 → 0.1.12

package/skills/yuanflow-skill//346/234/254/345/234/260/351/237/263/350/247/206/351/242/221/350/275/254/346/226/207/345/255/227/scripts/common/utils.py

@@ -0,0 +1,62 @@
+from __future__ import annotations
+
+import json
+import re
+from pathlib import Path
+from typing import Iterable
+
+
+VIDEO_SUFFIXES = {".mp4", ".mov", ".mkv", ".avi", ".webm", ".m4v"}
+AUDIO_SUFFIXES = {".wav", ".mp3", ".m4a", ".aac", ".flac", ".ogg", ".opus"}
+MEDIA_SUFFIXES = VIDEO_SUFFIXES | AUDIO_SUFFIXES
+INVALID_FILENAME_RE = re.compile('[<>:"/\\\\|?*\\x00-\\x1F]')
+
+
+def sanitize_filename(name: str, max_length: int = 120) -> str:
+    cleaned = INVALID_FILENAME_RE.sub("_", name).strip()
+    cleaned = re.sub(r"\s+", " ", cleaned)
+    cleaned = cleaned.rstrip(". ")
+    if not cleaned:
+        cleaned = "untitled"
+    return cleaned[:max_length].rstrip(". ")
+
+
+def ensure_parent(path: Path) -> Path:
+    path.parent.mkdir(parents=True, exist_ok=True)
+    return path
+
+
+def ensure_dir(path: Path) -> Path:
+    path.mkdir(parents=True, exist_ok=True)
+    return path
+
+
+def write_json(path: Path, data: object) -> Path:
+    ensure_parent(path)
+    path.write_text(json.dumps(data, ensure_ascii=False, indent=2), encoding="utf-8")
+    return path
+
+
+def write_text(path: Path, text: str) -> Path:
+    ensure_parent(path)
+    path.write_text(text, encoding="utf-8")
+    return path
+
+
+def iter_media_files(path: Path, recursive: bool = False) -> Iterable[Path]:
+    if path.is_file():
+        if path.suffix.lower() in MEDIA_SUFFIXES:
+            yield path
+        return
+    iterator = path.rglob("*") if recursive else path.glob("*")
+    for candidate in iterator:
+        if candidate.is_file() and candidate.suffix.lower() in MEDIA_SUFFIXES:
+            yield candidate
+
+
+def is_video_file(path: Path) -> bool:
+    return path.suffix.lower() in VIDEO_SUFFIXES
+
+
+def is_audio_file(path: Path) -> bool:
+    return path.suffix.lower() in AUDIO_SUFFIXES

package/skills/yuanflow-skill//346/234/254/345/234/260/351/237/263/350/247/206/351/242/221/350/275/254/346/226/207/345/255/227/scripts/requirements-transcribe.txt

@@ -0,0 +1,2 @@
+funasr>=1.1.6
+modelscope>=1.18.1

package/skills/yuanflow-skill//346/234/254/345/234/260/351/237/263/350/247/206/351/242/221/350/275/254/346/226/207/345/255/227/scripts/transcribe_media.py

@@ -0,0 +1,126 @@
+from __future__ import annotations
+
+import argparse
+from pathlib import Path
+
+from common.media import extract_audio
+from common.sensevoice import build_model
+from common.sensevoice import clean_transcript
+from common.utils import ensure_dir
+from common.utils import is_audio_file
+from common.utils import is_video_file
+from common.utils import iter_media_files
+from common.utils import sanitize_filename
+from common.utils import write_text
+
+
+SCRIPT_DIR = Path(__file__).resolve().parent
+DEFAULT_CACHE_ROOT = SCRIPT_DIR / "cache"
+DEFAULT_MODELS_ROOT = SCRIPT_DIR / "models"
+
+
+def prepare_audio(
+    source: Path,
+    *,
+    audio_cache_dir: Path,
+    ffmpeg_bin: str,
+    overwrite: bool,
+) -> tuple[Path, str]:
+    if is_audio_file(source):
+        return source, "audio"
+    if not is_video_file(source):
+        raise ValueError(f"不支持的媒体文件类型：{source}")
+    output_path = audio_cache_dir / f"{sanitize_filename(source.stem)}.wav"
+    extract_audio(
+        source,
+        output_path,
+        ffmpeg_bin=ffmpeg_bin,
+        overwrite=overwrite,
+    )
+    return output_path, "video"
+
+
+def transcribe_file(
+    model,
+    audio_file: Path,
+    *,
+    output_dir: Path,
+    language: str,
+    batch_size_s: int,
+    overwrite: bool,
+) -> Path:
+    output_path = output_dir / f"{sanitize_filename(audio_file.stem)}.txt"
+    if output_path.exists() and not overwrite:
+        return output_path
+    result = model.generate(
+        input=str(audio_file),
+        cache={},
+        language=language,
+        use_itn=True,
+        batch_size_s=batch_size_s,
+        merge_vad=True,
+        merge_length_s=15,
+        ban_emo_unk=True,
+    )
+    raw_text = ""
+    if result and isinstance(result, list):
+        raw_text = str(result[0].get("text", ""))
+    return write_text(output_path, clean_transcript(raw_text))
+
+
+def main() -> None:
+    parser = argparse.ArgumentParser(description="本地音视频转文字：视频先抽音频，音频直接转写。")
+    parser.add_argument("input_path", type=Path, help="音频文件、视频文件或目录")
+    parser.add_argument("--cache-root", type=Path, default=DEFAULT_CACHE_ROOT, help="缓存目录")
+    parser.add_argument("--models-root", type=Path, default=DEFAULT_MODELS_ROOT, help="模型目录")
+    parser.add_argument("--recursive", action="store_true", help="目录模式下递归扫描")
+    parser.add_argument("--overwrite", action="store_true", help="覆盖已有缓存和文本结果")
+    parser.add_argument("--ffmpeg-bin", default="ffmpeg", help="ffmpeg 可执行文件名或路径")
+    parser.add_argument("--device", default="auto", help="auto、cpu、cuda:0 等")
+    parser.add_argument("--language", default="auto", help="zh、en、yue、ja、ko、auto")
+    parser.add_argument("--batch-size-s", type=int, default=60, help="动态 batch 秒数")
+    args = parser.parse_args()
+
+    cache_root = args.cache_root.resolve()
+    models_root = args.models_root.resolve()
+    audio_cache_dir = ensure_dir(cache_root / "audio")
+    transcript_dir = ensure_dir(cache_root / "transcripts")
+
+    media_files = list(iter_media_files(args.input_path.resolve(), recursive=args.recursive))
+    if not media_files:
+        raise SystemExit("没有找到可处理的音频或视频文件。")
+
+    try:
+        model, downloaded_models = build_model(models_root, device=args.device)
+    except ImportError as exc:
+        raise SystemExit(
+            "缺少 FunASR 或 ModelScope 依赖。请先在 scripts/.venv 中安装 requirements-transcribe.txt。"
+        ) from exc
+
+    print(f"模型目录: {models_root}")
+    print(f"缓存目录: {cache_root}")
+    print(f"模型首次下载: {'是' if downloaded_models else '否'}")
+
+    for media_file in media_files:
+        audio_file, source_type = prepare_audio(
+            media_file,
+            audio_cache_dir=audio_cache_dir,
+            ffmpeg_bin=args.ffmpeg_bin,
+            overwrite=args.overwrite,
+        )
+        text_path = transcribe_file(
+            model,
+            audio_file,
+            output_dir=transcript_dir,
+            language=args.language,
+            batch_size_s=max(1, args.batch_size_s),
+            overwrite=args.overwrite,
+        )
+        print(f"输入文件: {media_file}")
+        print(f"输入类型: {'视频，已先抽取音频' if source_type == 'video' else '音频'}")
+        print(f"音频文件: {audio_file}")
+        print(f"转写结果: {text_path}")
+
+
+if __name__ == "__main__":
+    main()