Skip to content

README.md

Latest commit

 

History

History
192 lines (149 loc) · 8.2 KB

README.md

File metadata and controls

192 lines (149 loc) · 8.2 KB
Raw

Transcribix — Pinokio

1-click install and launch for Transcribix — offline speech-to-text transcription with 11 local AI models.

Pinokio Python License

What This Does

This is a Pinokio app that installs and runs Transcribix with a single click. It handles:

  • GPU detection — auto-installs CUDA or CPU PyTorch
  • Model group selection — install only what you need
  • Isolated venvs — each model group has its own environment
  • Gradio Web UI — browser-based interface for transcription

Installation

Option 1: Via Pinokio (Recommended)

  1. Open Pinokio
  2. Click "Discover" → search for "Transcribix"
  3. Click "Install" and select a model group
  4. Click "Start" to launch the Gradio Web UI

Option 2: Manual

git clone https://github.com/C0m3b4ck/Transcribix-Pinokio.git
cd Transcribix-Pinokio

# Install dependencies (pick one group)
pip install -r requirements/whisper.txt      # Whisper variants
pip install -r requirements/nvidia.txt       # NVIDIA models
pip install -r requirements/lightweight.txt  # Lightweight models
pip install -r requirements/all.txt          # All models
pip install -r requirements/gradio.txt       # Web UI

# Install PyTorch (auto-detects GPU)
python torch.js  # or manually: pip install torch torchaudio

# Launch
python app.py

Then open http://localhost:7860.

Model Groups

Group Models VRAM Disk Best For
Whisper faster-whisper, WhisperX, stable-ts, Distil-Whisper, Whisper, whisper.cpp 2-10 GB ~8 GB General use, subtitles
NVIDIA Parakeet TDT, Canary Qwen 2-6 GB ~5 GB Best accuracy
Lightweight Moonshine, SenseVoice, Vosk <1 GB ~2 GB Low资源 / edge devices
All All 11 models 2-10 GB ~20 GB Full comparison

Features

  • 11 local speech-to-text models — no API keys, no internet
  • Word-level timestamps for caption generation
  • SRT, VTT, and ASS subtitles with full styling (font, color, size, position, outline, shadow)
  • Optional subtitle burning into video via ffmpeg
  • Interactive file selection (upload or enter path)
  • GPU auto-detection (CUDA/CPU)
  • Colorful CLI with system info and position preview

Gradio Web UI

The web UI provides:

  • File upload or file path input — use either method
  • Model selection with live info display
  • Subtitle styling — font, color, size, position, outline, shadow
  • Words per caption — 1-8 words per subtitle block
  • Optional video burning — checkbox to burn subtitles into video (requires ffmpeg)
  • Downloadable outputs — SRT, ASS, and burned video files
┌─────────────────────────────────────────────────────────────┐
│  Transcribix                                                │
│  Offline speech-to-text with 11 local AI models             │
│                                                             │
│  Device: CUDA: NVIDIA RTX 3060 (12.0 GB VRAM)              │
│                                                             │
│  ┌─ Input ──────────────┐  ┌─ Subtitle Styling ──────────┐ │
│  │ [Upload File] [Path] │  │ Font: Arial    Size: 24     │ │
│  │ Model: Faster Whisper│  │ Color: White   Pos: Bottom  │ │
│  │ Language: en         │  │ Outline: 2     Shadow: 1    │ │
│  │ [x] Burn into video  │  │ Words/chunk: 3              │ │
│  └──────────────────────┘  └─────────────────────────────┘ │
│                                                             │
│  [  Transcribe & Generate Subtitles  ]                      │
│                                                             │
│  **Transcription complete!**                                │
│  - Model: Faster Whisper                                    │
│  - Words detected: 1234                                     │
│  - Time: 45.2s                                             │
│                                                             │
│  [SRT Subtitles]  [ASS Subtitles]  [Video with Subtitles]  │
└─────────────────────────────────────────────────────────────┘

Supported Formats

Audio Video
WAV, MP3, M4A, FLAC, OGG MP4, MKV, AVI, MOV, WebM

Project Structure

Transcribix-Pinokio/
├── pinokio.json               # Pinokio app metadata
├── pinokio.js                 # Pinokio menu system
├── app.py                     # Gradio web UI
├── captioning_comparison.py   # Main engine (11 models)
├── install.js                 # Group-aware installer
├── start.js                   # Launcher
├── torch.js                   # GPU-aware PyTorch installer
├── reset.js                   # Reset venv
├── resetgroup.js              # Reset specific group
├── update.js                  # Update dependencies
├── requirements/
│   ├── whisper.txt            # Whisper variants
│   ├── nvidia.txt             # NVIDIA models
│   ├── lightweight.txt        # Lightweight models
│   ├── all.txt                # All models
│   └── gradio.txt             # Gradio UI
└── venvs/                     # Isolated venvs (created on install)
    ├── whisper/
    ├── nvidia/
    ├── lightweight/
    └── all/

How It Works

  1. Install — Pinokio creates an isolated venv for your chosen model group
  2. GPU Detection — torch.js checks for CUDA and installs the right PyTorch
  3. Launch — start.js runs app.py and opens the Gradio Web UI
  4. Transcribe — Select a file, pick a model, customize subtitles, click Transcribe
  5. Output — Download SRT/ASS files or a video with burned-in subtitles

Requirements

Component Minimum Recommended
Python 3.10+ 3.11
RAM 8 GB 16 GB+
GPU VRAM 1 GB (Moonshine) 6 GB+ (large-v3)
Disk 2 GB 10 GB+
OS Linux, macOS, Windows Linux

Troubleshooting

"No module named gradio" → Install gradio: pip install -r requirements/gradio.txt

"ffmpeg not found" → Install ffmpeg: sudo apt install ffmpeg (Linux) or brew install ffmpeg (macOS)

"CUDA out of memory" → Use a smaller model (tiny/base) or switch to CPU mode

Pinokio won't start → Make sure you have Python 3.10+ installed and on your PATH

Credits

This project wraps Transcribix by C0m3b4ck.

Speech-to-Text Engines

Platforms

License

Apache License 2.0