import sys,subprocess,numpy as np
from faster_whisper import WhisperModel
p,a,b=sys.argv[1],float(sys.argv[2]),float(sys.argv[3]); raw=subprocess.check_output(['ffmpeg','-v','error','-ss',str(a),'-t',str(b-a),'-i',p,'-f','f32le','-ar','16000','-ac','1','-'])
m=WhisperModel('small',device='cpu',compute_type='int8')
s,info=m.transcribe(np.frombuffer(raw,dtype=np.float32),language='en',vad_filter=False,word_timestamps=True)
for x in s: print(a+x.start,a+x.end,x.text,flush=True)

import pathlib,subprocess,numpy as np,sys,json
r=pathlib.Path(__file__).resolve().parents[1];p=pathlib.Path(sys.argv[1]) if len(sys.argv)>1 else r/'output/离别_Farewell.mp4'
a=np.frombuffer(subprocess.check_output(['ffmpeg','-v','error','-i',str(p),'-vn','-ar','48000','-ac','1','-f','f32le','-']),np.float32)
total=float(subprocess.check_output(['ffprobe','-v','error','-show_entries','format=duration','-of','csv=p=0',str(p)]))
for lo,hi in [(total-3,total-2.5),(total-2,total-1.5),(total-1,total-.5),(total-.4,total-.1)]:
 b=a[round(lo*48000):round(hi*48000)];rms=float(np.sqrt(np.mean(b*b)));print(lo,hi,rms);assert rms>1e-5,f'BGM lost before fade ends at {lo}s'
print('PASS: music remains audible until the end of the fade')

{
  "width": 1280,
  "height": 720,
  "fps": 30,
  "transition": 0.65,
  "bgm": "At Rest - Kevin MacLeod.mp3",
  "bgm_start": 44,
  "bgm_delay": 3.5,
  "end_fade": 3,
  "segments": [
    {
      "source": "01-casablanca.mp4",
      "film": "Casablanca",
      "year": 1942,
      "inpoint": 44.8,
      "duration": 8.5,
      "cues": [
        {
          "start": 0.55,
          "end": 2.15,
          "zh": "那我们呢？",
          "en": "But what about us?",
          "kind": "dialogue"
        },
        {
          "start": 2.65,
          "end": 4.35,
          "zh": "我们永远拥有巴黎。",
          "en": "We'll always have Paris.",
          "kind": "dialogue"
        }
      ],
      "voices": [
        [
          0.45,
          4.25
        ]
      ],
      "crop": "652:480:98:0"
    },
    {
      "source": "02-romanholiday.mp4",
      "film": "Roman Holiday",
      "year": 1953,
      "inpoint": 154,
      "duration": 8,
      "cues": [
        {
          "start": 0.8,
          "end": 7.2,
          "zh": "我会珍藏在这里的回忆，直至生命尽头。",
          "en": "I will cherish my visit here in memory as long as I live.",
          "kind": "dialogue"
        }
      ],
      "voices": [
        [
          0.65,
          7.2
        ]
      ],
      "crop": null
    },
    {
      "source": "06-beforesunrise.mp4",
      "film": "Before Sunrise",
      "year": 1995,
      "inpoint": 124.8,
      "duration": 8,
      "cues": [
        {
          "start": 1.5,
          "end": 5.8,
          "zh": "再见。",
          "en": "Goodbye.",
          "kind": "dialogue"
        }
      ],
      "voices": [
        [
          1.7,
          7.05
        ]
      ],
      "crop": "iw:ih*0.84:0:0"
    },
    {
      "source": "08-lostintranslation.mp4",
      "film": "Lost in Translation",
      "year": 2003,
      "inpoint": 160,
      "duration": 8,
      "cues": [],
      "voices": [],
      "crop": null
    },
    {
      "source": "09-brokeback.mp4",
      "film": "Brokeback Mountain",
      "year": 2005,
      "inpoint": 236.7,
      "duration": 8,
      "cues": [
        {
          "start": 1.1,
          "end": 4.9,
          "zh": "我真希望知道，该如何戒掉你。",
          "en": "I wish I knew how to quit you.",
          "kind": "dialogue"
        }
      ],
      "voices": [
        [
          1.2,
          4.8
        ]
      ],
      "crop": "1920:1040:0:20"
    },
    {
      "source": "11-lalaland.mp4",
      "film": "La La Land",
      "year": 2016,
      "inpoint": 543,
      "duration": 8,
      "cues": [],
      "voices": [],
      "crop": "1920:752:0:164"
    },
    {
      "source": "10-interstellar.mp4",
      "film": "Interstellar",
      "year": 2014,
      "inpoint": 189.25,
      "duration": 9,
      "cues": [
        {
          "start": 1.7,
          "end": 3,
          "zh": "我爱你。",
          "en": "I love you.",
          "kind": "dialogue"
        },
        {
          "start": 6.4,
          "end": 8.2,
          "zh": "我会回来的。",
          "en": "And I'm coming back.",
          "kind": "dialogue"
        }
      ],
      "voices": [
        [
          1.65,
          2.75
        ],
        [
          6.4,
          8.25
        ]
      ],
      "crop": null
    },
    {
      "source": "07-titanic.mp4",
      "film": "Titanic",
      "year": 1997,
      "inpoint": 281.8,
      "duration": 8,
      "cues": [
        {
          "start": 2.1,
          "end": 3.45,
          "zh": "我永不放手。",
          "en": "I'll never let go.",
          "kind": "dialogue"
        },
        {
          "start": 3.5,
          "end": 5.4,
          "zh": "我保证。",
          "en": "I promise.",
          "kind": "dialogue"
        }
      ],
      "voices": [
        [
          2,
          4.5
        ]
      ],
      "crop": "1920:816:0:132"
    },
    {
      "source": "05-forrestgump.mp4",
      "film": "Forrest Gump",
      "year": 1994,
      "inpoint": 168.5,
      "duration": 8,
      "cues": [
        {
          "start": 1.1,
          "end": 4.5,
          "zh": "我想你，珍妮。",
          "en": "I miss you, Jenny.",
          "kind": "dialogue"
        }
      ],
      "voices": [
        [
          1.25,
          3.25
        ]
      ],
      "crop": null
    },
    {
      "source": "04b-cinemaparadiso.mp4",
      "film": "Cinema Paradiso",
      "year": 1988,
      "inpoint": 26,
      "duration": 8,
      "cues": [
        {
          "start": 3.2,
          "end": 6.9,
          "zh": "无论做什么，都要热爱它。",
          "en": "Whatever you end up doing, love it.",
          "kind": "dialogue"
        }
      ],
      "voices": [
        [
          3,
          7.2
        ]
      ],
      "crop": "iw:ih*0.70:0:0"
    },
    {
      "source": "12-goodwillhunting.mp4",
      "film": "Good Will Hunting",
      "year": 1997,
      "inpoint": 1,
      "duration": 8,
      "cues": [
        {
          "start": 3.1,
          "end": 6.25,
          "zh": "抱歉，我得去找一个姑娘。",
          "en": "Sorry, I had to go see about a girl.",
          "kind": "dialogue"
        }
      ],
      "voices": [
        [
          3,
          6.15
        ]
      ],
      "crop": "iw*0.76:ih*0.65:iw*0.12:ih*0.12"
    },
    {
      "source": "03-et.mp4",
      "film": "E.T.",
      "year": 1982,
      "inpoint": 88,
      "duration": 8,
      "cues": [],
      "voices": [],
      "crop": "1280:688:0:16"
    },
    {
      "source": "13-toystory.mp4",
      "film": "Toy Story 3",
      "year": 2010,
      "inpoint": 40.5,
      "duration": 9,
      "cues": [
        {
          "start": 2.1,
          "end": 4.7,
          "zh": "再见，老伙计。",
          "en": "So long, partner.",
          "kind": "dialogue"
        }
      ],
      "voices": [
        [
          2.2,
          4.4
        ]
      ],
      "crop": null
    }
  ],
  "framing": "cover"
}

#!/usr/bin/env python3
"""Portable, config-driven movie montage. Run: python build.py [segments|voices|join|titles|mix|all]."""
import pathlib,json,subprocess,sys,math
import numpy as np
from PIL import Image,ImageDraw,ImageFont
R=pathlib.Path(__file__).resolve().parents[1]; WK=R/'work'; C=json.loads((WK/'timeline.json').read_text()); S=C['segments']; W,H,FPS=C['width'],C['height'],C['fps']; XF=round(C['transition']*FPS)/FPS
for d in ['segments','overlays','logs']: (WK/d).mkdir(exist_ok=True)
clock=0
for i,s in enumerate(S):
 s['duration']=round(s['duration']*FPS)/FPS;s['start']=clock;clock+=s['duration']-(XF if i<len(S)-1 else 0)
TOTAL=round(clock*FPS)/FPS;FRAMES=round(TOTAL*FPS)
def run(args,name):
 print(name,flush=True)
 with open(WK/'logs'/(name+'.log'),'w') as log: subprocess.run(['ffmpeg','-hide_banner','-v','warning','-y',*map(str,args)],stdout=log,stderr=log,check=True)
def enc():return ['-c:v','libx264','-preset','fast','-crf','18','-pix_fmt','yuv420p','-threads','2']
def segments():
 for i,s in enumerate(S):
  dst=WK/'segments'/f'{i:02}.mp4'
  filt=[]
  if s.get('crop'):filt+=['crop='+s['crop']]
  filt+=[f'scale={W}:{H}:force_original_aspect_ratio=increase:flags=lanczos',f'crop={W}:{H}:(iw-ow)/2:(ih-oh)/2','setsar=1',f'fps={FPS}','settb=AVTB','setpts=PTS-STARTPTS','format=yuv420p']
  run(['-ss',s['inpoint'],'-i',R/'clips'/s['source'],'-t',s['duration'],'-an','-vf',','.join(filt),*enc(),'-frames:v',round(s['duration']*FPS),dst],f'segment-{i:02}')
def voices():
 sr=48000; master=np.zeros((round(TOTAL*sr),2),np.float32)
 for i,s in enumerate(S):
  if not s['voices']:continue
  raw=WK/'segments'/f'voice-raw-{i:02}.wav';stem=WK/'separated'/'htdemucs'/raw.stem/'vocals.wav'
  if not stem.exists():
   run(['-ss',s['inpoint'],'-i',R/'clips'/s['source'],'-t',s['duration'],'-vn','-ar','44100','-ac','2',raw],f'voice-extract-{i:02}')
   print('demucs',i,flush=True)
   with open(WK/'logs'/f'demucs-{i:02}.log','w') as log:subprocess.run([sys.executable,'-m','demucs','--two-stems','vocals','-n','htdemucs','--shifts','0','-d','cpu','-o',str(WK/'separated'),str(raw)],stdout=log,stderr=log,check=True)
  pcm=subprocess.check_output(['ffmpeg','-v','error','-i',str(stem),'-af','highpass=f=100,lowpass=f=8000,loudnorm=I=-18:TP=-3:LRA=7','-f','f32le','-ar',str(sr),'-ac','2','-'])
  a=np.frombuffer(pcm,np.float32).reshape(-1,2)
  for v0,v1 in s['voices']:
   l,h=round(v0*sr),min(round(v1*sr),len(a));b=a[l:h].copy();n=min(round(.04*sr),len(b)//2);b[:n]*=np.linspace(0,1,n)[:,None];b[-n:]*=np.linspace(1,0,n)[:,None]
   pos=round((s['start']+v0)*sr);master[pos:pos+len(b)]+=b
  print('voice isolated',s['film'],flush=True)
 p=subprocess.run(['ffmpeg','-v','error','-y','-f','f32le','-ar',str(sr),'-ac','2','-i','-','-c:a','pcm_s24le',str(WK/'voices.wav')],input=master.tobytes(),check=True)
def join():
 args=[]
 for i in range(len(S)):args+=['-i',WK/'segments'/f'{i:02}.mp4']
 filters=[];prev='0:v'
 for i in range(1,len(S)):
  lab=f'x{i}';filters.append(f'[{prev}][{i}:v]xfade=transition=fade:duration={XF}:offset={S[i]["start"]:.6f}[{lab}]');prev=lab
 run([*args,'-filter_complex_threads','1','-filter_complex',';'.join(filters),'-map',f'[{prev}]',*enc(),'-frames:v',FRAMES,WK/'picture.mp4'],'join')
def font(size,english=False):return ImageFont.truetype('/System/Library/Fonts/Supplemental/Arial.ttf' if english else '/System/Library/Fonts/Supplemental/Songti.ttc',size)
def textcenter(d,y,text,f,fill):
 box=d.textbbox((0,0),text,font=f);x=(W-(box[2]-box[0]))/2;d.text((x,y),text,font=f,fill=fill,stroke_width=2,stroke_fill=(0,0,0,230))
def overlay_img(s,cue=None):
 im=Image.new('RGBA',(W,H));d=ImageDraw.Draw(im)
 if cue is None:
  title=f'{s["film"]} {s["year"]}';ft=font(26,True);box=d.textbbox((0,0),title,font=ft)
  tw,th=box[2]-box[0],box[3]-box[1]
  d.rectangle((24,20,24+tw+24,20+th+20),fill=(0,0,0,255))
  d.text((36-box[0],30-box[1]),title,font=ft,fill=(255,255,255,255))
 if cue:
  textcenter(d,610,cue['zh'],font(31),(250,250,250,255));textcenter(d,656,cue['en'],font(21,True),(206,206,206,255))
 return im

def titles():
 # One finite alpha movie prevents dozens of looped PNG inputs and never overlaps title cards.
 events=[];srt=[];k=1
 for i,s in enumerate(S):
  lo=s['start']+(XF+.10 if i else 0);hi=s['start']+s['duration']-(XF+.10 if i<len(S)-1 else .15)
  events.append((lo,hi,overlay_img(s)))
  for q in s['cues']:
   a=max(lo,s['start']+q['start']);b=min(hi,s['start']+q['end']);events.append((a,b,overlay_img(s,q)))
   def tc(x):
    n=round(x*1000);return f'{n//3600000:02}:{n//60000%60:02}:{n//1000%60:02},{n%1000:03}'
   prefix=''
   srt.append(f'{k}\n{tc(a)} --> {tc(b)}\n{prefix}{q["zh"]}\n{q["en"]}\n');k+=1
 (R/'output'/'离别_中英字幕.srt').write_text('\n'.join(srt),encoding='utf-8')
 blank=Image.new('RGBA',(W,H));blankbytes=blank.tobytes();cache={};last=None
 with open(WK/'logs'/'overlay.log','w') as log:
  p=subprocess.Popen(['ffmpeg','-v','warning','-y','-f','rawvideo','-pixel_format','rgba','-video_size',f'{W}x{H}','-framerate',str(FPS),'-i','-','-an','-c:v','qtrle','-pix_fmt','argb','-frames:v',str(FRAMES),str(WK/'overlays.mov')],stdin=subprocess.PIPE,stderr=log)
  try:
   for f in range(FRAMES):
    t=f/FPS;active=[e for e in events if e[0]<=t<e[1]]
    if active:
     opacities=[max(0,min(1,(t-a)/.12,(b-t)/.12)) for a,b,im in active]
     key=tuple((id(e[2]),round(alpha*12)) for e,alpha in zip(active,opacities))
     if key not in cache:
      frame=blank.copy()
      for (a,b,im),opacity in zip(active,opacities):
       layer=im.copy();layer.putalpha(im.getchannel('A').point(lambda z:int(z*opacity)))
       frame=Image.alpha_composite(frame,layer)
      cache[key]=frame.tobytes()
     data=cache[key]
    else:data=blankbytes
    p.stdin.write(data)
   p.stdin.close();assert p.wait()==0
  except: p.kill();raise
 run(['-i',WK/'picture.mp4','-i',WK/'overlays.mov','-filter_complex_threads','1','-filter_complex',f'[0:v][1:v]overlay=shortest=1,fade=t=out:st={TOTAL-C["end_fade"]}:d={C["end_fade"]}[v]','-map','[v]',*enc(),'-frames:v',FRAMES,WK/'captioned.mp4'],'titles')
def mix():
 delay=C['bgm_delay'];fade=C['end_fade']
 filt=f'[0:a]atrim=duration={TOTAL},asetpts=PTS-STARTPTS,asplit=2[voice][side];[1:a]atrim=start={C["bgm_start"]}:duration={TOTAL-delay},asetpts=PTS-STARTPTS,loudnorm=I=-21:TP=-4:LRA=9,afade=t=in:d=2,adelay={round(delay*1000)}|{round(delay*1000)},asetpts=N/SR/TB,apad,atrim=duration={TOTAL}[music];[music][side]sidechaincompress=threshold=0.025:ratio=5:attack=35:release=450[duck];[duck][voice]amix=inputs=2:normalize=0,alimiter=limit=0.92:level=false,afade=t=out:st={TOTAL-fade}:d={fade}[mix]'
 run(['-i',WK/'voices.wav','-i',R/'bgm'/C['bgm'],'-i',WK/'captioned.mp4','-filter_complex',filt,'-map','2:v:0','-map','[mix]','-c:v','copy','-c:a','aac','-b:a','256k','-ar','48000','-t',TOTAL,'-movflags','+faststart',R/'output'/'离别_Farewell.mp4'],'mix')
 (WK/'resolved_timeline.json').write_text(json.dumps(dict(total=TOTAL,transition=XF,segments=S),ensure_ascii=False,indent=2))
 print('FINISHED',TOTAL,flush=True)
if __name__=='__main__':
 stage=sys.argv[1] if len(sys.argv)>1 else 'all'
 for name,func in [('segments',segments),('voices',voices),('join',join),('titles',titles),('mix',mix)]:
  if stage in ['all',name]:func()

import json,pathlib,subprocess,numpy as np
from PIL import Image,ImageDraw,ImageFont
r=pathlib.Path(__file__).resolve().parents[1];c=json.loads((r/'work/timeline.json').read_text());d=json.loads((r/'work/resolved_timeline.json').read_text());out=r/'output/离别_Farewell.mp4';report={}
info=json.loads(subprocess.check_output(['ffprobe','-v','error','-show_streams','-show_format','-of','json',str(out)]));report['duration']=float(info['format']['duration']);report['size']=int(info['format']['size']);report['expected_duration']=d['total'];report['streams']=[{k:s.get(k) for k in ['codec_name','width','height','r_frame_rate','sample_rate','channels']} for s in info['streams']]
assert abs(report['duration']-d['total'])<.08
assert all(5<=s['duration']<=10 for s in d['segments'])
font=ImageFont.truetype('/System/Library/Fonts/Supplemental/Arial.ttf',16)
points=[(s['start']+min(3.3,s['duration']/2),s['film']) for s in d['segments']]
points += [(s['start']+d['transition']/2,'TRANSITION '+str(i)) for i,s in enumerate(d['segments'][1:],1)]
im=Image.new('RGB',(1280,mathrows:=((len(points)+3)//4)*205))
for i,(t,label) in enumerate(points):
 p=r/'work'/f'qa-{i:02}.jpg';subprocess.run(['ffmpeg','-v','error','-y','-ss',str(t),'-i',str(out),'-frames:v','1','-vf','scale=320:180',str(p)],check=True)
 im.paste(Image.open(p),(i%4*320,i//4*205));ImageDraw.Draw(im).text((i%4*320+4,i//4*205+182),f'{t:.2f} {label}',font=font,fill='white')
im.save(r/'output/逐段与转场核验.jpg',quality=92)
blacks=[]
for t in [d['total']-3,d['total']-1.5,d['total']-.1,d['total']-1/30]:
 b=subprocess.check_output(['ffmpeg','-v','error','-ss',str(t),'-i',str(out),'-frames:v','1','-pix_fmt','rgb24','-f','rawvideo','-']);a=np.frombuffer(b,np.uint8);blacks.append({'time':t,'mean':float(a.mean()),'max':int(a.max())})
report['end_black']=blacks;assert blacks[-1]['mean']<2
raw=subprocess.check_output(['ffmpeg','-v','error','-i',str(out),'-vn','-ac','2','-ar','48000','-f','f32le','-']);a=np.frombuffer(raw,np.float32).reshape(-1,2);report['audio_peak']=float(np.abs(a).max());report['audio_last_100ms_rms']=float(np.sqrt((a[-4800:]**2).mean()));report['one_second_rms']=[float(np.sqrt((a[i:i+48000]**2).mean())) for i in range(0,len(a),48000)];assert report['audio_peak']<1
for s in d['segments']:
 s['subtitles_verified']=all(s['start']+q['end']<=s['start']+s['duration'] for q in s['cues'])
report['segment_count']=len(d['segments']);report['voice_segments']=[s['film'] for s in d['segments'] if s['voices']];report['status']='passed technical checks; contact sheet requires visual review'
(r/'output/验收报告.json').write_text(json.dumps(report,ensure_ascii=False,indent=2));print(json.dumps(report,ensure_ascii=False,indent=2))