subtitles

Generate JP and EN subtitles

---

How does this work?

1. Extract vocals from input for better transcription

Using Mel-Band-Roformer.

• https://github.com/KimberleyJensen/Mel-Band-Roformer-Vocal-Model

2. Generate Transcription

Using a variant of OpenAI Whisper.

• https://github.com/Purfview/whisper-standalone-win

3. Translate to English

Using Google Gemini.

• https://github.com/MaKTaiL/gemini-srt-translator

---

How to use the Colab file?

1. Choose an Input Method

Option A - Download via yt-dlp

Option B - Google Drive

• Upload the video to Google Drive.
• Then specify the path to the file in Google Drive.
• Keep it simple. No spaces or special characters. No effort was made to accomodate those.

2. Choose an Output folder in your Google Drive

3. Input Google Gemini API key(s)

You need at least 1 key. If desired, you can specify 2 keys from different accounts to try to get around Google free API limits.

Option A - Just fill in the keys.

Option B - Use Colab's Secret (Better)

Put your keys in Colab's Secret by clicking the key icon. Make sure that Notebook access is enabled. Key names are GEMINI_API_KEY_1 and GEMINI_API_KEY_2.

4. Run

A. Make sure that you enable GPU.

B. Run.