Vesta

Warning

Work in Progress: This README is currently temporary and a work in progress (WIP), is subject to ongoing reorganization, and will be further refined and expanded.

subs2srs, but actually fast.

Vesta is a modern desktop application for language learners and power users that turns video and subtitle files into rich, synchronized Anki flashcard decks, auto-aligned subtitles, and translated media in minutes instead of hours — running ~2.5× faster than subs2srs with parallel multi-core processing.

---

Core Features

1. Flashcard Generation & Anki (.apkg / TSV) Export

• Self-Contained .apkg Packages: Exports native SQLite Anki collections with zero manual import mapping needed.
• Rich Media Cards: Attach synchronized audio snippets, high-resolution snapshots, or compact video clips to each card.
• Modern Codec Support:
  • Audio: MP3 or Opus (ultra-low bitrate speech compression).
  • Snapshots: WebP, AVIF, or JPEG with customizable resolution presets (144p to 1080p) and quality tuning.
  • Video Clips: H.264 or MPEG-4 with hardware acceleration and ultrafast encoding presets.
• Audio & Video Enhancements:
  • EBU R128 Loudness Normalization: Balances whispering and loud action scenes across cards.
  • Audio Track Selection: Extract from multi-language audio streams (e.g. Japanese audio from dual-audio releases).
  • Subtitle Crop: Automatically crop bottom borders to remove burned-in hardsubs from snapshots.
  • Context & Sentence Merging: Attach leading/trailing context dialogue lines or automatically join split subtitle sentences.
• Card Styling & Dark Mode: Beautiful, responsive card templates with native dark-mode support and automatic font stack injection tailored for target languages (CJK Noto, Arabic, Thai, Devanagari, Hebrew, Cyrillic, etc.).

2. Difficulty Tagging & Vocabulary Profiling

Vesta automatically analyzes the lexical complexity of each subtitle sentence and tags cards with their proficiency level (e.g., HSK::3, CEFR::B1, JLPT::N2, TOPIK::3, TOCFL::B2):

Built-in Databases & Provenance:

• 🇨🇳 HSK (Simplified Chinese) — 12,543 entries (HSK 1–6): Official Hanban / CTI syllabus & Ministry of Education PRC with Simplified/Traditional variants (gigacover/hsk).
• 🇹🇼 TOCFL (Traditional Chinese) — 11,170 entries (Levels 1–6 / Novice to Superior): Official Taiwan Ministry of Education SC-TOP 8000 Vocabulary List (Downloads).
• 🇯🇵 JLPT (Japanese) — 13,930 entries (Levels N5–N1): Jonathan Waller's JLPT (Tanos) & EDRDG / JMdict (Kanji, Kana, and dictionary forms).
• 🇰🇷 TOPIK (Korean) — 6,671 entries (Levels 1–6): Official NIIED & National Institute of Korean Language standard vocabulary syllabus.
• 🇬🇧 CEFR English — 8,845 entries (Levels A1–C2): Cambridge English Profile (EVP) & Oxford 3000/5000 aligned with the Council of Europe CEFR.
• 🇪🇺 CEFR Multi-lingual European Databases (🇩🇪 German 41,900, 🇮🇹 Italian 19,951, 🇪🇸 Spanish 19,945, 🇫🇷 French 19,941, 🇵🇹 Portuguese 19,950, 🇷🇺 Russian 19,970 entries): Frequency distributions derived from OPUS OpenSubtitles (HermitDave) and Leipzig Corpora Collection, mapped to 6 logarithmic CEFR brackets (A1=Top 1K to C2=20K+).

📖 Full Linguistic Provenance & Reproduction Guide: See docs/VOCABULARY_SOURCES.md for full dataset licenses, exact entry counts, methodology, and the deterministic build script.

• Zero Internet Required & 1-Click Database Export: Embedded directly into the binary. Export any database to .tsv with one click from Settings for inspection, customization, or community sharing.
• Custom Vocabulary TSVs & User Schemes: Load your own custom frequency or vocabulary lists with user-defined tag prefixes.
• Configurable Unknown Word Policies:
  • Ignore: Ignore unlisted words and tag based on known vocabulary.
  • Highest: Treat unlisted/rare words as maximum difficulty.
  • Level 0: Assign explicit Level 0 (e.g. HSK::0, CEFR::0, JLPT::0, TOPIK::0, TOCFL::0, Level::0) for unclassified sentences.

3. Speech-to-Text Transcription

Generate accurate SRT subtitles directly from media files:

• Local Whisper (whisper.cpp): Offline transcription with GPU acceleration (Vulkan) and beam search quality modes.
• Silero VAD (Voice Activity Detection): Pre-filters silence and background music, dramatically reducing hallucinations and subtitle drift.
• Cloud STT Providers: Integrated support for Groq, OpenAI, Mistral, and Deepgram for lightning-fast cloud transcription.

4. Smart Synchronization & Alignment

• Anchor-based Re-timing (srt-sync): Align drifting subtitles interactively using waveform anchors.
• Automatic Whisper Re-sync (srt-autosync): Automatically generate phonetic anchors with Whisper to realign out-of-sync subtitles with zero manual effort.

5. AI Subtitle Translation (srt-translate)

• Translate foreign subtitle lines into your native language using Large Language Models.
• Context-aware batching to preserve dialogue flow, slang, and pronouns.
• Multi-tier provider failover (Ollama local, OpenAI, Claude, DeepSeek, OpenRouter).

6. Smart Episode & Subtitle Matching

• Drag-and-drop video and subtitle files in bulk.
• Automatically pairs files across complex naming conventions:
  • Western TV: S01E05, 1x05, Episode 01, Folge 06, Episodio 03, Серия 09.
  • Anime / Fansub releases: [Group] Title - 01 [1080p].mkv, OVA 01, SP 02.
  • Chinese / Korean dramas: 第01话, 第12集, 01화.
• Auto-detects original vs reference subtitle roles (source, vostfr, sub_ita, traduzione, etc.).

7. Additional Modules & Integrations

• Dialogue Condenser (srt-condense): Strips silence and non-speech intervals to generate condensed audio for listening immersion.
• AnkiConnect Direct Sync (srt-ankiconnect): Push notes, media, and decks directly into a running Anki instance without manual .apkg file import.
• Deck Refiner (srt-refine): Enrich existing Anki decks using LLMs with explanations, grammar notes, and usage examples.

---

Pipeline Overview

Video / Audio ──► [Transcribe] ──► [Sync / Autosync] ──► [Translate] ──► [Flashcards & Anki Export]

Already have subtitles? Skip directly to Sync or Flashcards generation.

---

Modular Architecture & Headless CLI Use

Vesta is organized as a Cargo workspace of decoupled, single-responsibility crates. Every backend engine is a GUI-agnostic library in lib/ or core/ paired with a standalone CLI binary in cli/.

vesta/
├── core/
│   ├── srt-parser/        # SRT parsing, timing, and formatting engine
│   └── srt-apkg/          # Native Anki package (.apkg) SQLite collection builder
├── lib/
│   ├── srt-flashcards/    # Flashcard generation, media orchestration & filters
│   ├── srt-difficulty/    # Lexical complexity analyzer & multi-lingual vocab tables
│   ├── srt-transcribe/    # Whisper.cpp + Silero VAD + Cloud STT pipeline
│   ├── srt-autosync/      # Automatic Whisper-assisted subtitle synchronization
│   ├── srt-sync/          # Anchor-based timing interpolation
│   ├── srt-translate/     # Multi-tier LLM subtitle translation
│   ├── srt-condense/      # Dialogue extraction and audio condensation
│   ├── srt-ankiconnect/   # AnkiConnect API client for direct sync
│   ├── srt-extract/       # Subtitle text and metadata extraction
│   └── srt-refine/        # LLM-powered deck enrichment
├── cli/
│   ├── srt-flashcards-cli # Headless flashcard & deck generation CLI
│   ├── srt-transcribe-cli # Media to SRT transcription CLI
│   ├── srt-autosync-cli   # Subtitle re-sync CLI
│   ├── srt-translate-cli  # Subtitle translation CLI
│   └── srt-extract-cli    # Subtitle data extraction CLI
└── apps/
    ├── srt-gui/           # Desktop GUI (Tauri + Svelte 5 + Tailwind)
    └── whisper-bench/     # Transcription benchmarking tool

CLI Quick Examples

Build only the standalone CLI tool you need:

# Build the flashcards CLI
cargo build --release -p srt-flashcards-cli

# Generate an Anki .apkg deck from subtitles and video
./target/release/srt-flashcards generate \
  --target episode-ja.srt \
  --native episode-en.srt \
  --video episode.mkv \
  --output ./out_deck \
  --format apkg \
  --deck "Anime Season 1" \
  --snapshot-format webp \
  --audio-format opus

For comprehensive module guides and Rust integration examples, see docs/modules/README.md and docs/ARCHITECTURE.md.

---

Documentation Map

• docs/ARCHITECTURE.md — Architectural design contracts, layering rules, and conventions.
• docs/VOCABULARY_SOURCES.md — Exact linguistic dataset provenance, upstream sources, licenses, and build scripts.
• docs/modules/ — Detailed module specifications and embedding instructions.
• docs/superpowers/specs/ — Technical specifications for media presets, codec evaluations, and format benchmarks.
• benchmarking_against_subs2srs/ — Reproducible benchmarking scripts and methodology.

---

Building from Source

Prerequisites

• Rust: 1.85+ (rustup default stable)
• Node.js: 18+ and npm
• System dependencies: ffmpeg and ffprobe on your system PATH.

Development Setup

# Clone the repository
git clone https://github.com/pierspad/vesta.git
cd vesta

# Install frontend dependencies
cd apps/srt-gui && npm install && cd ../..

# Run GUI in development mode
./run_gui.sh
# Or manually:
cd apps/srt-gui && npx tauri dev

---

Contributing

Pull requests are welcome! For major changes or architectural proposals, please open an issue first to discuss what you would like to change.

---

License & Disclosures

• License: GNU General Public License v3.0 (GPLv3) — see the LICENSE file for details.
• AI Disclosure: This project was developed with the assistance of Large Language Models to support implementation, refactoring, and documentation.