AI Subtitle Generator & Auto-Sync Engine

01. Overview

The Subtitle Generation Module uses OpenAI's open-source Whisper Series (Fast-Whisper implementation) as its recognition core. It can automatically transcribe any major video or audio file (such as MP4, MKV, MP3, M4A, WAV, FLAC, AAC, OGG) into accurate, time-stamped subtitles (SRT/VTT).

This system not only supports basic subtitle generation but also integrates Auto-Translation, Bilingual Subtitles layout, and a one-click Auto Burn-in feature, allowing you to produce finished, subtitled videos without opening other editing software.

02. AI Core Parameters Explained

🧠 AI Model Selection

We provide four different tiers of models. You can choose based on your PC's performance and accuracy needs:

• medium: Fast VRAM: ~2GB
  Suitable for videos with clear, standard dialogue. Fastest speed, but slightly weaker at recognizing proper nouns or fast speech.
• large-v2 (Good): Balanced VRAM: ~4GB
  Currently the most stable version, striking a great balance between accuracy and speed. Suitable for most YouTube videos.
• large-v3-turbo (Better): New Tech VRAM: ~6GB
  A speed-optimized V3 version, retaining V3's high-recognition traits but computing faster.
• large-v3 (Best): Highest Quality VRAM: ~10GB+
  OpenAI's latest and most powerful model. Most resistant to multi-language mixing, accents, and background noise, but requires a powerful graphics card and plenty of VRAM.

🗣️ Video Language

Specify the language spoken in the video. While setting it to Auto is usually accurate, if the video has very sparse dialogue or loud background music, manually specifying the language (e.g., English) can significantly improve accuracy.

⚙️ Processing Mode

• Auto-Translate (to Traditional Chinese): Regardless of whether the video is English, Japanese, or Korean, the system will automatically translate and output "Traditional Chinese" subtitles.
• Original (Original Text): The language spoken in the video dictates the output subtitle language (e.g., Japanese video outputs Japanese subtitles).

🛡️ VAD (Voice Activity Detection) Filter

This is a highly important feature. If checked, the AI will first analyze "where there is human voice" before transcribing.
Purpose: Prevents the AI from hallucinating subtitles (like meaningless symbols or repeated words) during purely musical or silent segments.
Recommendation: Turn on manually depending on the situation. It is turned off by default in the latest version to prevent accidentally deleting soft-spoken or whispered dialogue.

👥 Speaker Diarization

Uses Pyannote AI for audio track separation, automatically recognizing how many people are speaking in the video, and pre-pending speaker tags (e.g., [SPEAKER_00]:, [SPEAKER_01]:) to the beginning of subtitles.

1. Go to the pyannote/speaker-diarization-3.1 and pyannote/segmentation-3.0 model pages.
2. Ensure you are logged into your Hugging Face account.
3. If you haven't agreed to the terms, an agreement and form will appear. Fill it out and click "Agree and access repository".
4. If you click in and directly see the model description and "Files and versions" list, it means you have successfully authorized!

03. Subtitle Output & Style

📄 Output Formats

• SRT: The most universal subtitle format, containing timestamps and text.
• VTT (WebVTT): Web-specific subtitle format, supporting simple styling.
• TXT: Plain text file, no timestamps. Ideal for organizing transcripts.
• Auto Burn-in: "Burns" the subtitles directly onto the video frames, generating a new MP4 file.

🎨 Independent Style Settings (Exclusive Lossless ASS Dual Engine)

Our subtitle generation core utilizes a powerful Dual-Layer Overlay Dynamic Calculation Engine. It not only guarantees compatibility with all video resolutions but also allows the "Primary Subtitle" and "Secondary Subtitle (Original Text)" to have completely independent visual designs without interfering with each other!

On the right side of the interface, we provide an intuitive Tabview:

• Primary Layout: Controls the primary translated language (usually Chinese).
• Secondary Layout (Original): Controls the secondary language underneath (only visible when "Bilingual Subtitles" is checked).

Within each tab, you can independently configure the following parameters, and the settings will sync in real-time to the Preview above:

• Font & Bold: Click the button to select from all system-installed fonts (like Microsoft JhengHei).
• Color & Size: Customize your text color freely.
• Spacing: Customize the distance between characters in the subtitle (supports values from 0-30) for more flexible layout formatting.
• Outline: Independently thicken the stroke around the text to prevent glare.
• Background Box: Add a semi-transparent custom background color behind the text (for the ultimate Netflix-style viewing experience). Through underlying PrimaryStyle / SecondaryStyle calculation tech, even if you overlap two lines with different colored boxes, the layout will never break!