~/.openclaw/venvs/sensevoice/bin/python3 -c "
from funasr import AutoModel
from funasr.utils.postprocess_utils import rich_transcription_postprocess
from datetime import datetime, timedelta
import re, sys
wav = '<音频文件路径>'
out = '<输出路径>.txt'
# 尝试从文件名解析录音开始时间
m = re.search(r'(\d{8})_(\d{6})', wav)
start_dt = datetime.strptime(m.group(1)+m.group(2), '%Y%m%d%H%M%S') if m else None
vad_model = AutoModel(model='fsmn-vad', disable_update=True)
model = AutoModel(model='iic/SenseVoiceSmall', vad_model='fsmn-vad',
vad_kwargs={'max_single_segment_time': 30000}, device='cpu')
vad_segs = vad_model.generate(input=wav)[0].get('value', [])
res = model.generate(input=wav, cache={}, language='zh', use_itn=True,
batch_size_s=60, merge_vad=True)
texts = [rich_transcription_postprocess(s).strip()
for s in re.split(r'<\|zh\|>', res[0]['text']) if s.strip()]
texts = [s for s in texts if len(s) > 1]
# 情绪标签是模型产物，转写结果中需清理
emoji_pat = re.compile(r'[\U0001F600-\U0001F64F\U0001F300-\U0001F5FF\U0001F680-\U0001F6FF\U0001F700-\U0001F77F\U0001F780-\U0001F7FF\U0001F800-\U0001F8FF\U0001F900-\U0001F9FF\U0001FA00-\U0001FA6F\U0001FA70-\U0001FAFF\U00002702-\U000027B0\U0000FE00-\U0000FE0F\U0000200D]+')
ratio = len(vad_segs) / len(texts) if texts else 1
lines = []
if start_dt:
lines.append(f'[录音开始: {start_dt.strftime(\"%H:%M:%S\")}]')
for i, t in enumerate(texts):
t = emoji_pat.sub('', t).strip()
if not t:
continue
vi = min(int(i * ratio), len(vad_segs)-1)
ts = (start_dt + timedelta(milliseconds=vad_segs[vi][0])).strftime('%H:%M:%S') if start_dt else f'{vad_segs[vi][0]//1000:.0f}s'
lines.append(f'[{ts}] {t}')
with open(out, 'w', encoding='utf-8') as f:
f.write('\n'.join(lines))
print(f'Done: {len(texts)} segments -> {out}')
"
