Skip to content

Releases: ventura8/Auto-Subtitle-Generator

v1.0.1

26 Jan 18:19
@ventura8 ventura8
v1.0.1
4d00b71

Choose a tag to compare

View all tags
v1.0.1 Latest
Latest

Release v1.0.1 - Runtime Fixes & Hardware Compatibility

This release focuses on improving the stability of the installation process and resolving critical compatibility issues with specific hardware, most notably the NVIDIA RTX 30x0 series (specifically the RTX 3080).

🚀 What's New

  • RTX 3080 Compatibility: Fixed a bug that caused model loading failures on newer NVIDIA cards by refining the device mapping and weight tying logic.
  • Improved Performance Profiles: Enhanced the behavior of ULTRA, HIGH, and MID profiles to ensure better resource allocation across CPU and GPU.
  • Performance Overrides: Users can now explicitly override Whisper beam size and NLLB batch sizes in config.yaml to fine-tune performance for their specific hardware.

🛠️ Bug Fixes

  • Windows File Lock Fix: Added a retry mechanism (3 attempts with delay) when cleaning up temporary files. This resolves the PermissionError: [WinError 32] often caused by Windows file locks during the cleanup phase.
  • Robust Installer: The install_dependencies.ps1 script has been updated with better error handling for pip installations and environment setup.
  • Configuration Persistence: Fixed an issue where manual overrides in config.yaml were sometimes ignored in favor of profile defaults.

📄 Documentation

  • Updated README.md and instructions.md with updated "Golden Standard" performance recommendations for different hardware tiers.
  • Added clearer instructions for manual dependency resolution.

💾 Installation

To update, simply pull the latest changes and run the installer again:

./install_dependencies.ps1

Full Changelog: v1.0.0...v1.0.1

Assets 2
Loading

v1.0.0

12 Jan 21:53
@ventura8 ventura8
v1.0.0
2609c20

Choose a tag to compare

View all tags
v1.0.0
AI Hybrid VHS Audio Restorer Logo

AI Hybrid VHS Audio Restorer

A specialized audio restoration pipeline designed to remaster VHS recordings.

The Pipeline

This project uses a hybrid AI approach to separate, clean, and remix audio:

  1. Separation: Uses audio-separator with the BS-Roformer-Viperx-1297 model to split audio into Vocals and Music/Instrumental stems with state-of-the-art accuracy.
  2. Vocal Enhancement: Runs Resemble-Enhance on the vocal track to remove muffle/hiss, restore high frequencies, and improve clarity.
  3. Music Denoising: Runs audio-separator with the UVR-DeNoise-Lite model on the music track to remove tape hiss while preserving musical details and preventing distortion.
  4. Mixing: Recombines the clean vocals and music using FFmpeg for a balanced, high-fidelity 32-bit PCM output.

Requirements

The installer handles everything, ensuring compatibility with modern hardware:

  • Python 3.10+ (in a local venv)
  • FFmpeg 6.1+ (Full Portable Build included & configured)
  • PyTorch 2.x Nightly (Required for RTX 50-series/Sm_120 support)
  • Audio-Separator (BS-Roformer & UVR-DeNoise models)
  • Resemble-Enhance (Windows-Patched for vocal restoration)
  • Runtime Patcher: Automatically fixes torchaudio incompatibility and deepspeed crashes on Windows.

Hardware Auto-Detection Logic

The script automatically scales performance based on your GPU VRAM:

Profile VRAM Example GPUs Batch Size
ULTRA ≥ 22 GB RTX 3090 / 4090 / 5090 16
HIGH ≥ 15 GB RTX 4080 / 5080 8
MID ≥ 10 GB RTX 3080 / 4070 4
LOW < 10 GB Entry Level / Older Cards 1

Note

CPU threads are automatically set to your maximum available cores (e.g., 32 threads for Ryzen 9950X3D).

Usage

  1. Run install_dependencies.ps1 to set up the environment.

Option A: Drag & Drop (Recommended)

Simply drag and drop your video file(s) or a folder containing videos directly onto start_workflow.bat (or the Python script).

  • Output: The restored video will be saved in the same folder as your original video.

Option B: Interactive Mode (Default)

Double-click start_workflow.bat without any files.

  • The script will Launch and show your System Stats.
  • Press Enter to automatically scan and process all files in the input folder.
  • Output: The restored videos will be saved in the output folder.

Option C: CLI

Run via command line with arguments:

python restore_audio_hybrid.py "C:\Path\To\Video.mp4"
  • Output: The restored video will be saved in the same folder as the input video.
  1. Find the restored videos in output.

Credits

Loading