Show HN: Open-source subtitle generation for seamless content translation (github.com)

A developer released an open-source, self-hosted subtitle-generation tool that uses whisper.cpp for high-performance on-device inference, enabling automated, multilingual subtitle creation and optional merging into video containers. It’s designed for privacy-conscious users and teams who want full control over processing: the repo includes a setup script that clones and compiles whisper.cpp (auto-detecting macOS/Linux and Metal/CUDA support), installs a whisper-cli binary, and downloads default models. The tool uses Conda for environment management and requires standard video-processing tools (git, make, cmake, ffmpeg). Usage is simple: python subtitle.py <video_or_URL> [--model <name>] produces .vtt files in a data/ folder and can merge subtitles where supported. This matters for the AI/ML community because it packages a practical, reproducible pipeline for subtitle generation that balances accuracy, latency, and privacy via locally run models. Supported models range from tiny/tiny.en up to large-v3 (plus English-only .en variants), letting users trade speed for quality. The project is easy to integrate into workflows, useful for content localization, accessibility, and research into speech-to-text deployment. For developers it offers an approachable base to extend or benchmark whisper.cpp-powered transcription in production-like, self-hosted environments.