videocaptioner

VideoCaptioner is a video subtitle processing tool based on large language models, covering the entire workflow of speech recognition, subtitle optimization, translation, and video synthesis. Installation is via pip install videocaptioner, with both CLI and GUI versions available. Core features include: - Speech transcription: Supports multiple engines such as faster-whisper, whisper-api, bijian (free), jianying (free), and whisper-cpp, using word-level timestamps and VAD (Voice Activity Detection) to improve recognition accuracy. - Subtitle optimization and translation: Leverages LLMs for semantic understanding and sentence segmentation, making subtitles read naturally and fluently; translation supports LLM, bing (free), google (free) and other services, with context-aware translation and reflection-based optimization. - Video synthesis: Supports burning subtitles into videos as soft or hard subtitles. - Dubbing generation: Can generate dubbed audio tracks or dubbed videos from subtitles. - Online video download: Supports platforms like YouTube and Bilibili. - Full workflow processing: Complete transcription, optimization, translation, and synthesis in a single command. Free features (Bijian speech recognition, Bing/Google translation) require no configuration. LLM features (subtitle optimization, large model translation) require an API Key and support all OpenAI-compatible interfaces, including VideoCaptioner relay service, SiliconCloud, DeepSeek, etc. Configuration priority is: CLI parameters > environment variables > config file > defaults. Additionally, the project provides a Claude Code Skill, allowing AI coding assistants to directly invoke VideoCaptioner for video processing. Development uses uv for dependency management, with pyright type checking and pytest testing. The project is licensed under GPL-3.0.