Video Translation

Video Translation Tool — User Guide

Translate Guinness Global Investors video content into multiple languages using AI-powered dubbing, voice cloning, and quality review.

Quick Start

The typical workflow follows these steps:

1. Transcribe — Fetch videos from YouTube and generate English transcripts using AI speech recognition.
2. Voices — Detect speakers in each video and clone their voices for use in translations.
3. Translate — Select a video and target language. The system translates text, generates speech, and lip-syncs the result automatically.
4. Review — AI back-translates the dubbed audio and flags potential errors. Fix issues, re-dub, then accept.
5. Publish — Optionally upscale video quality, then upload to YouTube.

1. Transcribe

Fetches videos from the Guinness GI YouTube channel and creates English transcripts.

How to use

• Click Refresh to sync the latest videos from YouTube.
• Select one or more videos, then click Transcribe Selected.
• Transcription runs in the background — you can navigate away.
• Once complete, review and Accept or Reject each transcript.

Search & Sort

• Use the search box to filter videos by title.
• Sort by most recent, most viewed, or alphabetically.

2. Voices

Detect and clone speaker voices so translated audio sounds like the original presenter.

How to use

• The left column shows videos needing voice assignment; the right shows completed ones.
• Select a video and click Auto-detect Voices to identify speakers using audio embeddings.
• For each detected speaker, review the audio sample and confidence score.
• Click Clone Voice to create a voice clone, or match to an existing cloned voice.
• Click Accept when all speakers are assigned.

3. Translate

One-click translation from English into a target language.

Supported languages

Chinese, German, Spanish, French, Italian, Portuguese, Japanese

How to use

• Select a transcribed video with voices assigned.
• Choose a target language.
• Click Translate. The system will:
  1. Translate all segments
  2. Generate speech for each segment
  3. Time-stretch audio to fit original segment durations
  4. Lip-sync the speaker's mouth to the new audio
  5. Composite the final dubbed video
• Monitor progress in the active jobs bar at the top of the page.

4. Dubbing Studio

Fine-tune translations segment by segment before generating the dubbed video.

How to use

• Select a video and language from the sidebar.
• Navigate through segments using Previous / Next buttons.
• Edit translated text directly in the text field.
• Click Generate Dubbed Video when you're happy with the translations.

5. Review

Quality assurance for completed translations. The system back-translates the dubbed content and uses AI to flag potential errors.

How to use

• Filter by language using the sub-tabs (All, Chinese, German, etc.).
• Click a flag icon to open the QA review panel for that translation.
• The quality summary shows an overall rating: Good, Acceptable, or Poor.

Reviewing segments

The segment table shows each chunk with:

• Original English — the source transcript text
• Translated text — what was sent to TTS
• Back-translation — the translated text converted back to English
• Quality flag — OK, Warning, Error, or Fixed

Fixing issues

• Fix button (on flagged segments) — AI corrects the translation automatically.
• Edit original text — Click the English text cell to fix transcription errors. After saving, click Re-translate to update the translation.
• Re-dub — After fixing segments, click Re-dub Fixed Segments to regenerate audio. Only changed segments are re-processed.

Accepting or rejecting

• Accept — Marks the translation as approved and ready for publishing.
• Reject — Removes the translation. The transcript and voice assignments are kept so you can re-translate.

6. Upscale

Enhance video resolution with AI upscaling (optional step before publishing).

Models

• Proteus — Best all-around quality
• Iris — Optimised for face enhancement (recommended for presenter videos)
• Artemis — Denoise + sharpen
• Gaia — Animation/CG content

How to use

• Select a completed translation video.
• Choose a model and scale factor (2x or 4x).
• Click Upscale and monitor progress.

7. Chat

Ask questions about video content using AI. Useful for quickly finding what's discussed across the video library.

How to use

• Click a video card to ask questions about that specific video.
• Click the Jess avatar for general questions across all videos.
• Responses include clickable timestamps to jump to relevant sections.

8. Pipeline

Batch process multiple videos through the full translation workflow in one go.

How to use

• Review the pipeline summary to see how many videos are at each stage.
• Select videos and target languages.
• Click Process. The system will automatically:
  1. Transcribe videos (if needed)
  2. Auto-match speaker voices
  3. Translate + dub for each selected language
• Already-completed items are skipped automatically.
• Monitor progress in the active jobs section.

Tips & Troubleshooting

Common issues

• "Video file not available" — The dubbed video file is missing from the server. Click Re-dub Now in the Review panel to regenerate.
• QA review shows "Unknown" quality — Back-translation failed (usually a temporary API issue). Re-open the review and it will retry automatically.
• Translation stuck at "Processing" — Check the Debug tab for errors. The job may have failed silently. You can re-submit the translation.
• Wrong words in transcript — The AI transcription sometimes mishears words. Click the English text in the Review panel to edit, then Re-translate.

Best practices

• Always review translations in the Review tab before publishing.
• Fix flagged segments before re-dubbing — only changed segments get new audio, so it's fast.
• Use the Pipeline tab for bulk work; use individual tabs for corrections.
• Check the Debug tab if something seems stuck — it shows API status and recent errors.

Access levels

• Viewer — Can browse videos, view transcripts, chat, and see translations (read-only).
• Creator — Can transcribe, translate, edit, clone voices, approve translations, and upscale.