Video-subtitle-extractor is a GUI tool for extracting hard-coded subtitles from videos and generating SRT files.
The tool solves the problem of recovering subtitle text that has been burned directly into video frames, which cannot be extracted through standard subtitle track access. It works by identifying key frames from video files, detecting text regions within those frames using deep learning-based object detection, performing optical character recognition to extract the text content, filtering out non-subtitle elements like watermarks and channel logos, removing duplicate lines, and outputting the results as SRT subtitle files or plain text. The extraction process operates entirely locally without requiring external API calls or online OCR services.
The tool suits projects where you need to recover subtitles from videos with hard-coded text and want to avoid dependency on third-party services. It supports batch processing of multiple videos simultaneously, provided they share consistent resolution and subtitle positioning. The tool offers three extraction modes: fast mode using lightweight models for quick results with potential minor text loss, automatic mode that selects between lightweight and precise models based on available hardware, and precise mode for frame-by-frame detection with maximum accuracy but significantly slower processing. It handles subtitles in multiple languages including Simplified Chinese, Traditional Chinese, English, Japanese, Korean, Vietnamese, Arabic, French, German, Russian, Spanish, Portuguese, Italian, and others. GPU acceleration is available to improve both accuracy and extraction speed. The tool includes a configuration file for mapping and removing specific text patterns, such as replacing common OCR errors or deleting recurring watermarks.
Development activity shows consistent engagement with user feedback through issue tracking and discussion channels. The project maintains active support for multiple operating systems and Python versions. Regular refinement of the extraction pipeline suggests ongoing optimization of the detection and recognition components. The tool's multi-mode approach indicates iterative development responding to real-world performance requirements across different hardware configurations.
