Subtitles

Transcription → Agent proofreading → Manual review → Burn-in

Core Flow (~8-15 minutes total, including manual review)

1. Extract audio + upload          ~1min
    ↓
2. Volcengine transcription (with hot words)  ~2min
    ↓
3. Agent auto-proofread            ~3-5min
    ↓
4. Manual review & confirm         Depends on user
    ↓
5. Burn-in subtitles               ~1-2min

---

Step 1: Extract Audio and Upload

# Extract audio
ffmpeg -i "video.mp4" -vn -acodec libmp3lame -y audio.mp3

# Upload to uguu.se (temporary file hosting)
curl -s -F "files[]=@audio.mp3" https://uguu.se/upload
# Returns URL like: https://o.uguu.se/xxxxx.mp3

---

Step 2: Volcengine Transcription (with Hot Words)

The transcription script automatically reads the dictionary as hot words to improve accuracy:

# Dictionary location: <project>/.claude/skills/qcut-toolkit/videocut/subtitles/dictionary.txt
# Script loads it automatically

bash ../talk-edit/scripts/volcengine_transcribe.sh "https://o.uguu.se/xxxxx.mp3"

Dictionary format (one word per line):

skills
Claude
Agent

---

Step 3: Agent Auto-Proofread

3.1 Generate Timestamped Subtitles

const result = JSON.parse(fs.readFileSync('volcengine_result.json'));
const subtitles = result.utterances.map((u, i) => ({
  id: i + 1,
  text: u.text,
  start: u.start_time / 1000,
  end: u.end_time / 1000
}));
fs.writeFileSync('subtitles_with_time.json', JSON.stringify(subtitles, null, 2));

3.2 Agent Manual Proofread (No Scripts)

After transcription, Agent must read all subtitles line by line and manually proofread.

Common misrecognition patterns vary by language. Agent should:

1. Check for homophones and similar-sounding word errors
2. Check for missing words in sentences
3. Verify proper nouns and technical terms against the dictionary
4. Fix grammar errors

3.3 Cross-Reference with Script (If Available)

If an original script exists, use it as reference but do not auto-match (text differences cause timestamp drift).

Agent should:

1. Read the script as reference
2. Manually compare line by line
3. Mark uncertain items for manual review

---

Step 4: Start Review Server

cd subtitles-dir/
node <project>/.claude/skills/qcut-toolkit/videocut/subtitles/scripts/subtitle_server.js 8898 "video.mp4"

Visit http://localhost:8898

Features:

• Left: video playback, Right: subtitle list
• Auto-highlight current subtitle during playback
• Double-click subtitle text to edit (timestamps preserved)
• Variable playback speed (1x/1.5x/2x/3x)
• Save subtitles / Export SRT / Burn-in subtitles
• Dictionary quick-insert at bottom

---

Step 5: Burn-in Subtitles

Default style: Size 22, golden bold, black outline 2px, bottom center

ffmpeg -i "video.mp4" \
  -vf "subtitles='video.srt':force_style='FontSize=22,FontName=PingFang SC,Bold=1,PrimaryColour=&H0000deff,OutlineColour=&H00000000,Outline=2,Alignment=2,MarginV=30'" \
  -c:a copy \
  -y "video_subtitled.mp4"

Parameter	Value	Description
FontSize	22	Font size
FontName	PingFang SC	PingFang font
Bold	1	Bold
PrimaryColour	&H0000deff	Golden #ffde00
OutlineColour	&H00000000	Black outline
Outline	2	Outline width
Alignment	2	Bottom center
MarginV	30	Bottom margin

---

Directory Structure

output/YYYY-MM-DD_video-name/subtitles/
├── 1_transcription/
│   ├── audio.mp3
│   └── volcengine_result.json
├── subtitles_with_time.json    # Core file
└── 3_output/
    ├── video.srt
    └── video_subtitled.mp4

---

Subtitle Rules

Rule	Description
One line per screen	No line breaks, no stacking
No end punctuation	Hello not Hello.
Keep mid-sentence punctuation	Click here, then there