| Stage | What you do | Typical tools | Output |
|-------|-------------|---------------|--------|
| A. Get the source video | Download/locate the MP4 (or any container) that is ~2 h 18 m long. | Any media player, wget, youtube‑dl, etc. | AVOP‑249‑orig.mp4 |
| B. Generate a rough transcript | Use an automatic speech‑recognition (ASR) engine to produce a time‑coded draft. | Whisper (OpenAI), Vosk, AssemblyAI, Google Speech‑to‑Text, YouTube auto‑captions | draft.txt (or draft.srt with rough timestamps) |
| C. Refine & sync | Clean up wording, split/merge lines, adjust timings, add speaker tags, sound cues, etc. | Aegisub, Subtitle Edit, Jubler, Subtitle Workshop | Cleaned SRT/WEBVTT file |
| D. Quality‑check | Play video + subtitles, look for overlaps, missed words, and readability. | Any media player that supports external subtitles (VLC, MPC‑Hc, MPV). | Final AVOP‑249‑engsub.srt |
| E. Optional: Hard‑burn | Embed subtitles into the video (so they’re always visible). | ffmpeg (-vf subtitles=) or HandBrake. | AVOP‑249‑engsub‑burned.mp4 |
If you spot an issue, go back to Aegisub, adjust, and re‑export.
pip install -U openai-whisper
# Also need ffmpeg (for audio extraction)
# Windows: choco install ffmpeg | macOS: brew install ffmpeg | Linux: sudo apt-get install ffmpeg
# tiny model ≈ 2 × real‑time, large model ≈ 0.5 × real‑time
whisper "AVOP-249-orig.mp4" --model large --language en --output_format srt --output_dir ./transcripts
Result → AVOP-249-orig.srt (rough timestamps, ~2 h 18 m total). AVOP-249-engsub Convert02-18-14 Min
Tip: If you have a GPU, add
--device cuda.
Tip: For very long videos, split the file first (e.g., every 30 min) usingffmpeg -ss … -t ….
| Problem | Fix |
|---------|-----|
| Whisper crashes on a 2 h+ file | Split the video first: ffmpeg -i AVOP-249-orig.mp4 -ss 00:00:00 -t 01:00:00 part1.mp4 (repeat for each chunk). Then run Whisper on each chunk and later concatenate the SRTs (cat part*.srt > combined.srt). |
| Subtitles lag by ~0.5 s | In Aegisub, select all lines (Ctrl+A) → Timing → Shift Times → negative 500 ms. |
| Too many “[Music]” cues | Use a noise gate in Audacity to isolate background music and only add a cue where it’s prominent. |
| Exported SRT shows weird characters | Ensure your editor saves as UTF‑8 without BOM. In Aegisub: File → Save Subtitles As… → choose UTF‑8. |
| ffmpeg says “Subtitle codec not found” | You likely need the libass library. Install it (brew install libass or sudo apt-get install libass-dev) and re‑run ffmpeg. | | Stage | What you do | Typical
Follow the steps above, and you’ll have a clean, time‑accurate English subtitle file for the 2 h 18 min “AVOP‑249‑engsub Convert02‑18‑14 Min” video in under an hour. If you hit any snags, just drop a follow‑up question—happy subtitling!
Below is a tiny excerpt that shows the correct formatting for a 2 h 18 m video. Use it as a template when you manually add or split lines. If you spot an issue, go back to
1
00:00:02,500 --> 00:00:05,800
[Music fades in]
2
00:00:06,000 --> 00:00:09,300
JANE: Hey, did you see the report from last night?
3
00:00:09,500 --> 00:00:12,200
MARK: Yeah, the numbers are… (cough) impressive.
4
00:00:12,400 --> 00:00:15,000
[Door slams]
5
00:00:15,200 --> 00:00:18,100
JANE: We need to act fast.
If you're dealing with such a file, here are a few general tips: