161 lines
5.7 KiB
Python
161 lines
5.7 KiB
Python
#!/usr/bin/env python3
|
||
"""
|
||
全自动删除视频中「嗯」的语音段落
|
||
使用 whisper-timestamped 词级时间戳检测 嗯，ffmpeg 裁剪输出
|
||
"""
|
||
import argparse
|
||
import json
|
||
import re
|
||
import subprocess
|
||
import tempfile
|
||
from pathlib import Path
|
||
|
||
# 语助词（词级匹配）
|
||
FILLER_CHARS = "嗯啊呃额哦噢唉哎诶喔"
|
||
|
||
|
||
def extract_audio(video_path: str, out_path: str) -> bool:
|
||
r = subprocess.run([
|
||
"ffmpeg", "-y", "-i", video_path, "-vn", "-acodec", "pcm_s16le", "-ar", "16000", out_path
|
||
], capture_output=True)
|
||
return r.returncode == 0
|
||
|
||
|
||
def get_duration(video_path: str) -> float:
|
||
r = subprocess.run(
|
||
["ffprobe", "-v", "error", "-show_entries", "format=duration", "-of", "default=noprint_wrappers=1:nokey=1", video_path],
|
||
capture_output=True, text=True
|
||
)
|
||
return float(r.stdout.strip()) if r.returncode == 0 else 0
|
||
|
||
|
||
def transcribe_word_level(audio_path: str, language: str = "zh") -> dict:
|
||
import whisper_timestamped as whisper
|
||
audio = whisper.load_audio(audio_path)
|
||
model = whisper.load_model("small", device="cpu") # small 更准，易识别语助词
|
||
# 用 initial_prompt 提示模型保留 嗯 等语助词
|
||
return whisper.transcribe(
|
||
model, audio, language=language, detect_disfluencies=True,
|
||
initial_prompt="嗯 啊 呃 然后 就是 那个 所以说。这是一段中文语音转写，请保留说话人发出的嗯、啊等语气词。"
|
||
)
|
||
|
||
|
||
def find_filler_ranges(result: dict) -> list:
|
||
"""从词级转录结果中找出语助词（嗯等）的时间段：词级 + 段级兜底"""
|
||
ranges = []
|
||
for seg in result.get("segments", []):
|
||
for w in seg.get("words", []):
|
||
text = (w.get("text") or "").strip()
|
||
t_clean = re.sub(r"[\s，。、,.\-—…]+", "", text)
|
||
if not t_clean:
|
||
continue
|
||
if re.match(r"^[嗯啊呃噢哦额唉哎诶喔]+$", t_clean):
|
||
s, e = w.get("start"), w.get("end")
|
||
if s is not None and e is not None and e - s > 0.05:
|
||
ranges.append((float(s), float(e)))
|
||
# 段级兜底：整段仅为语助词
|
||
seg_text = re.sub(r"[\s，。、,.\-—…]+", "", (seg.get("text") or "").strip())
|
||
if seg_text and re.match(r"^[嗯啊呃噢哦额唉哎诶喔]+$", seg_text):
|
||
ss, se = seg.get("start"), seg.get("end")
|
||
if ss is not None and se is not None:
|
||
r = (float(ss), float(se))
|
||
if r not in ranges and all(r[0] >= x[1] or r[1] <= x[0] for x in ranges):
|
||
ranges.append(r)
|
||
return sorted(ranges, key=lambda x: x[0])
|
||
|
||
|
||
def build_keep_ranges(remove_ranges: list, total_duration: float) -> list:
|
||
remove_ranges = sorted(remove_ranges, key=lambda x: x[0])
|
||
keep = []
|
||
current = 0.0
|
||
for rs, re in remove_ranges:
|
||
if rs > current + 0.05:
|
||
keep.append((current, rs))
|
||
current = max(current, re)
|
||
if current < total_duration - 0.05:
|
||
keep.append((current, total_duration))
|
||
return keep
|
||
|
||
|
||
def run_ffmpeg(args: list) -> bool:
|
||
return subprocess.run(args, capture_output=True).returncode == 0
|
||
|
||
|
||
def main():
|
||
ap = argparse.ArgumentParser()
|
||
ap.add_argument("video", help="输入视频")
|
||
ap.add_argument("-o", "--output", help="输出路径")
|
||
args = ap.parse_args()
|
||
|
||
video_path = Path(args.video).resolve()
|
||
if not video_path.exists():
|
||
print("❌ 视频不存在")
|
||
return 1
|
||
|
||
output_path = Path(args.output) if args.output else video_path.parent / f"{video_path.stem}_去嗯.mp4"
|
||
total_duration = get_duration(str(video_path))
|
||
if total_duration <= 0:
|
||
print("❌ 无法获取时长")
|
||
return 1
|
||
|
||
with tempfile.TemporaryDirectory() as tmpdir:
|
||
tmp = Path(tmpdir)
|
||
audio_path = tmp / "audio.wav"
|
||
print("1. 提取音频...")
|
||
if not extract_audio(str(video_path), str(audio_path)):
|
||
print("❌ 提取音频失败")
|
||
return 1
|
||
|
||
print("2. whisper-timestamped 词级转录（检测语助词）...")
|
||
try:
|
||
result = transcribe_word_level(str(audio_path))
|
||
except Exception as e:
|
||
print(f"❌ 转录失败: {e}")
|
||
return 1
|
||
|
||
remove_ranges = find_filler_ranges(result)
|
||
print(f" 检测到 {len(remove_ranges)} 处语助词（嗯等）")
|
||
for s, e in remove_ranges[:10]:
|
||
print(f" {s:.2f}s - {e:.2f}s")
|
||
if len(remove_ranges) > 10:
|
||
print(f" ... 共 {len(remove_ranges)} 处")
|
||
|
||
if not remove_ranges:
|
||
print(" 无检测到语助词，复制原视频作为输出")
|
||
import shutil
|
||
shutil.copy(str(video_path), str(output_path))
|
||
print(f"✅ 已复制: {output_path}")
|
||
return 0
|
||
|
||
keep_ranges = build_keep_ranges(remove_ranges, total_duration)
|
||
|
||
print("3. ffmpeg 裁剪并拼接...")
|
||
seg_files = []
|
||
for i, (start, end) in enumerate(keep_ranges):
|
||
dur = end - start
|
||
if dur < 0.1:
|
||
continue
|
||
seg = tmp / f"seg_{i:04d}.mp4"
|
||
if run_ffmpeg(["ffmpeg", "-y", "-ss", str(start), "-t", str(dur), "-i", str(video_path), "-c", "copy", str(seg)]):
|
||
seg_files.append(seg)
|
||
|
||
if not seg_files:
|
||
print("❌ 片段生成失败")
|
||
return 1
|
||
|
||
list_path = tmp / "list.txt"
|
||
with open(list_path, "w") as f:
|
||
for seg in seg_files:
|
||
f.write(f"file '{seg}'\n")
|
||
|
||
if not run_ffmpeg(["ffmpeg", "-y", "-f", "concat", "-safe", "0", "-i", str(list_path), "-c", "copy", str(output_path)]):
|
||
print("❌ 拼接失败")
|
||
return 1
|
||
|
||
print(f"✅ 已输出: {output_path}")
|
||
return 0
|
||
|
||
|
||
if __name__ == "__main__":
|
||
exit(main())
|