McCloudS/subgen

What it solves

Subgen is a self-hosted tool that generates high-accuracy subtitles (.srt or .lrc) for personal media libraries. It fills the gap when official subtitles are missing, out of sync, or unavailable, allowing users to create subtitles locally on their own hardware using AI transcription.

How it works

Subgen uses stable-ts and faster-whisper to transcribe audio/video files. It can transcribe non-English languages to themselves or translate foreign audio into English. The tool supports both CPU and Nvidia GPU (CUDA) acceleration and includes a specific fix for audio start-time offsets common in Amazon WEB-DL files to ensure timestamps are perfectly aligned.

Who it’s for

It is designed for home media server enthusiasts who use platforms like Bazarr, Plex, Emby, Jellyfin, or Tautulli. It integrates as a Whisper provider for Bazarr or via webhooks from media servers to automatically trigger transcription when media is added or played.

Highlights

• Broad Integration: Seamlessly connects with Bazarr, Plex, Emby, Jellyfin, and Tautulli.
• Hardware Acceleration: Supports CPU, Nvidia GPU (CUDA), and potentially ROCm for AMD GPUs.
• Flexible Model Selection: Supports various Whisper models, including large-v3-turbo for speed and distil-large-v3 for efficiency.
• Automatic Offset Correction: Detects and compensates for audio stream start-time offsets using ffprobe to prevent early subtitles.
• Folder Monitoring: Can watch specific folders for changes to automatically generate subtitles for new files.