JochenYang / shiyu

What is shiyu?

Shiyu is a desktop application that generates subtitles from audio and video files using local AI. Drop in a video, and it outputs SRT, LRC, or ASS subtitle files with accurate timing. The backend runs a Python FastAPI service with SenseVoice-Small for speech recognition, while the frontend is a Vue 3 app wrapped in Tauri for native desktop performance. Everything stays on your machine - no API keys, no cloud calls, no data leaving your computer.

Why is it gaining traction?

The privacy angle is the obvious hook - transcribe sensitive recordings without sending them anywhere. But the real differentiator is the timing accuracy. The -150ms latency compensation aligns subtitles to the audio waveform better than most cloud services, and the bilingual segmentation handles Chinese and English speech naturally. The model auto-downloads on first launch, and the UI is polished enough that it feels like a finished product rather than a weekend project. For developers who need subtitles for tutorials, meetings, or content creation, this fills a gap that previously required either paid services or clunky command-line tools.

Who should use this?

Content creators working with video tutorials who want to add subtitles without uploading to third-party services. Developers building applications that need offline speech-to-text capabilities. Researchers processing interview recordings or focus group data where confidentiality matters. Anyone frustrated with Whisper API costs or the friction of cloud-based transcription pipelines.

Verdict

Shiyu delivers on its core promise with a well-designed interface and genuinely useful offline AI. The credibility score of 0.949999988079071% reflects a very young project with only 10 stars - this is early-stage software with limited community validation. Documentation is functional but sparse, and there's no visible test suite. That said, the architecture is solid, the model download system is robust, and the codebase shows thoughtful engineering. Worth trying if you need local subtitle generation, but treat it as a promising v1 rather than a mature tool.