Technology

MioSub's subtitle generation engine and editor are built from scratch, optimized for AI-powered subtitle workflows.

Subtitle Generation Engine

Automatic Terminology Extraction

The problem with traditional machine translation: inconsistent proper noun translations — "東京" becomes "Tokyo" in one place and "Dongjing" in another.

MioSub's approach:

• Intelligently extract proper nouns from audio (names, places, titles, etc.)
• Verify standard translations via search engines
• Generate a glossary for consistent terminology throughout the video

Long-Context Translation

Traditional MT processes sentence by sentence, losing context and meaning.

MioSub's approach:

• Segment by semantics into 5-10 minute chunks
• Preserve full context for translation, understanding speaker intent
• Support scene presets (anime, film, news, tech) for optimized translation style

Post-Transcription Processing

Raw Whisper output has issues with sentence segmentation and timing accuracy.

MioSub's approach:

• Smart segmentation: Auto-split subtitles based on semantics and pauses
• Timeline correction: Fix timing drift in Whisper output
• Terminology replacement: Auto-apply glossary for consistent translations

CTC Forced Alignment

High-precision timeline alignment based on CTC (Connectionist Temporal Classification) technology.

• Millisecond-level character alignment
• Built-in aligner in v3.0, works out of the box
• Auto-downloads model on first use

Speaker Identification

Automatically identify and label speakers in multi-person conversations.

• LLM-based speaker inference
• Customizable speaker names and colors
• Merge adjacent subtitles from the same speaker

Smart Concurrency Control

Dynamically adjust concurrency based on model to maximize speed while avoiding rate limits:

Result: 30-minute video processed in ~8-10 minutes.

Fully Automated Pipeline

Paste a video link (YouTube/Bilibili), and the entire workflow runs automatically:

1. Auto Download — yt-dlp fetches the best quality video
2. Audio Extraction — Extract audio and perform VAD segmentation
3. Smart Transcription — Whisper speech-to-text
4. AI Translation & Polish — Gemini context-aware translation and refinement
5. Auto Hardcoding — FFmpeg burns bilingual subtitles (GPU acceleration supported)
6. Output — Ready-to-share MP4 with hardcoded subtitles

Subtitle Editor

Real-Time Preview

• Built-in assjs rendering engine for accurate font, color, and position rendering
• WYSIWYG — edit and preview simultaneously
• One-click toggle between source and translated text for quick review

Smart Caching

• Efficient transcoding cache for smooth playback
• Audio-only file support with adaptive player UI

Batch Operations

• Batch Regenerate: Re-run the full pipeline on selected segments (transcription → polish → align → translate)
• Polish Translation: Optimize translation quality for selected segments while maintaining context
• Auto-save snapshots before operations, rollback anytime

Version Management

• Auto-save edit history
• Snapshot rollback support
• Prevent accidental work loss