Making Korean subtitles for a YouTube video

Subtitles are made in Google Colab (free Google computers) with Whisper, the most accurate speech-to-text model. Your Mac only passes along a link, so it takes no space on it.

Once: set it up

1. Get the notebook. Download korean_subs.ipynb
2. Put it in Colab. Go to colab.research.google.com → ファイル → ノートブックをアップロード → pick the file. After that it lives in your Google Drive (folder Colab Notebooks) and opens from there.
3. Install the cookie extension. In Chrome, install Get cookies.txt LOCALLY from the Chrome Web Store.
4. Let it work in incognito. Type chrome://extensions in the address bar → Get cookies.txt LOCALLY → 詳細 → switch on シークレット モードでの実行を許可.
5. Export your YouTube cookies. Open a new incognito window (⌘ + Shift + N) → go to youtube.com → sign in (a spare Google account is safest) → while on the youtube.com tab, click the extension → Export. Close the incognito window right away.
6. Put the cookies in Drive. Rename the file to exactly cookies.txt and put it in Google Drive → マイドライブ → korean_subs (next to the subs folder, not inside it). The first run of the notebook creates that folder; you can also make it yourself.

Every time: make subtitles

1. Open the notebook in Colab.
2. ランタイム → ランタイムのタイプを変更 → L4 GPU → 保存. (A100 is faster but uses more units; T4 if you are low on units. The subtitles are identical.)
3. ランタイム → すべてのセルを実行. Allow Google Drive access when it asks.
4. A box appears under the second cell: paste a YouTube link, press Enter. Paste more links the same way. When done, press Enter on the empty box.
5. Wait for 🎉 all done. A 20-minute video takes a couple of minutes.
6. Google Drive → korean_subs → subs → select the new .srt files (⌘-click) → right-click → ダウンロード.
7. Here in Hana: Add subtitles → pick them. Don't rename the files — the [code] in the name is how Hana finds the video.

If something goes wrong

“Sign in to confirm you're not a bot”

The cookies expired. Delete the old cookies.txt from Drive and redo step 5 of the setup — signing in inside the incognito window.

“0 file(s) to transcribe”

Colab is looking at an old copy of your Drive. ランタイム → ランタイムを接続解除して削除, check the GPU is still L4, then run all again.

A subtitle file has repeated lines or gibberish

In the second cell change MODEL = 'large-v3' to MODEL = 'large-v2'. To redo a file: in Drive move its audio from done back to inbox, delete its old .srt from subs, run all, press Enter on the empty link box.

A video won't play in Hana

Some uploaders block playing outside YouTube. In the player click Open file from this computer and pick the audio from Drive → korean_subs → done (download it first).

Subtitles are a bit early or late

In the player use − / + next to the offset, or ⌃⇧← / ⌃⇧→.