143 lines
5.1 KiB
Python
143 lines
5.1 KiB
Python
#!/usr/bin/env python3
|
|
"""
|
|
批量 Whisper 转录脚本
|
|
- 并行度: 4 workers (medium模型, 31GB RAM安全)
|
|
- 输出: 每个视频的 .txt 逐字稿 + batch_summary.json
|
|
"""
|
|
import subprocess
|
|
import json
|
|
import os
|
|
import sys
|
|
import time
|
|
from concurrent.futures import ProcessPoolExecutor, as_completed
|
|
from pathlib import Path
|
|
|
|
VIDEO_DIR = "/root/.openclaw/workspace-xiaoban/tmp/Lina先生_视频/Lina先生_作品下载"
|
|
OUTPUT_DIR = "/root/.openclaw/workspace-xiaoban/tmp/whisper_output"
|
|
MODEL = "medium"
|
|
LANGUAGE = "Chinese"
|
|
WORKERS = 4 # 保守并行度
|
|
|
|
os.makedirs(OUTPUT_DIR, exist_ok=True)
|
|
|
|
def get_videos():
|
|
"""获取所有待处理视频列表"""
|
|
videos = []
|
|
for f in sorted(Path(VIDEO_DIR).glob("*.mp4")):
|
|
# 跳过已处理的
|
|
out_txt = Path(OUTPUT_DIR) / (f.stem + ".txt")
|
|
if out_txt.exists() and out_txt.stat().st_size > 0:
|
|
continue
|
|
# 获取时长
|
|
try:
|
|
result = subprocess.run(
|
|
["ffprobe", "-v", "error", "-show_entries", "format=duration",
|
|
"-of", "csv=p=0", str(f)],
|
|
capture_output=True, text=True, timeout=10
|
|
)
|
|
duration = float(result.stdout.strip()) if result.stdout.strip() else 0
|
|
except:
|
|
duration = 0
|
|
videos.append({
|
|
"path": str(f),
|
|
"name": f.name,
|
|
"duration": duration
|
|
})
|
|
return videos
|
|
|
|
def transcribe(video_info):
|
|
"""单个视频转录"""
|
|
f = video_info["path"]
|
|
name = video_info["name"]
|
|
duration = video_info["duration"]
|
|
out_txt = os.path.join(OUTPUT_DIR, Path(f).stem + ".txt")
|
|
|
|
# 如果已存在有效输出，跳过
|
|
if os.path.exists(out_txt) and os.path.getsize(out_txt) > 10:
|
|
with open(out_txt, 'r') as fh:
|
|
content = fh.read()
|
|
return {"name": name, "duration": duration, "text": content, "elapsed": 0, "cached": True}
|
|
|
|
start = time.time()
|
|
try:
|
|
result = subprocess.run(
|
|
["/usr/local/bin/whisper", f,
|
|
"--model", MODEL,
|
|
"--language", LANGUAGE,
|
|
"--output_dir", OUTPUT_DIR,
|
|
"--output_format", "txt"],
|
|
capture_output=True, text=True, timeout=1800 # 30min max per video
|
|
)
|
|
elapsed = time.time() - start
|
|
|
|
if os.path.exists(out_txt):
|
|
with open(out_txt, 'r') as fh:
|
|
content = fh.read()
|
|
else:
|
|
content = "[转录失败]"
|
|
|
|
return {"name": name, "duration": duration, "text": content, "elapsed": elapsed, "cached": False}
|
|
except subprocess.TimeoutExpired:
|
|
elapsed = time.time() - start
|
|
return {"name": name, "duration": duration, "text": "[超时]", "elapsed": elapsed, "cached": False}
|
|
except Exception as e:
|
|
elapsed = time.time() - start
|
|
return {"name": name, "duration": duration, "text": f"[错误: {e}]", "elapsed": elapsed, "cached": False}
|
|
|
|
def main():
|
|
videos = get_videos()
|
|
total = len(videos)
|
|
total_duration = sum(v["duration"] for v in videos)
|
|
|
|
print(f"待处理: {total} 个视频, 总时长: {total_duration/3600:.1f} 小时")
|
|
print(f"模型: {MODEL}, 并行: {WORKERS} workers")
|
|
print(f"预计耗时: {total_duration * 2.6 / WORKERS / 3600:.1f} 小时")
|
|
print("=" * 60)
|
|
|
|
done = 0
|
|
total_elapsed = 0
|
|
start_time = time.time()
|
|
|
|
with ProcessPoolExecutor(max_workers=WORKERS) as executor:
|
|
futures = {executor.submit(transcribe, v): v for v in videos}
|
|
|
|
for future in as_completed(futures):
|
|
v = futures[future]
|
|
done += 1
|
|
try:
|
|
result = future.result()
|
|
cached = result.get("cached", False)
|
|
elapsed = result.get("elapsed", 0)
|
|
total_elapsed += elapsed
|
|
|
|
pct = done / total * 100
|
|
elapsed_all = time.time() - start_time
|
|
eta = elapsed_all / done * (total - done) if done > 0 else 0
|
|
|
|
tag = "♻️缓存" if cached else f"⏱{elapsed:.0f}s"
|
|
print(f"[{done}/{total} {pct:.1f}%] {tag} | ETA: {eta/60:.0f}min | {result['name'][:50]}...")
|
|
|
|
# 每20个输出一次进度文件
|
|
if done % 20 == 0:
|
|
progress = {
|
|
"done": done, "total": total,
|
|
"elapsed_total": elapsed_all,
|
|
"eta_seconds": eta,
|
|
"pct": pct
|
|
}
|
|
with open(os.path.join(OUTPUT_DIR, "_progress.json"), 'w') as fp:
|
|
json.dump(progress, fp, indent=2)
|
|
|
|
except Exception as e:
|
|
print(f"[{done}/{total}] ❌ 错误: {v['name'][:50]}... => {e}")
|
|
|
|
total_time = time.time() - start_time
|
|
print(f"\n✅ 全部完成! 总耗时: {total_time/3600:.1f} 小时")
|
|
|
|
# 写入最终进度
|
|
with open(os.path.join(OUTPUT_DIR, "_progress.json"), 'w') as fp:
|
|
json.dump({"done": total, "total": total, "status": "complete"}, fp, indent=2)
|
|
|
|
if __name__ == "__main__":
|
|
main()
|