Original Summary

Originally, I am .NET software engineer working corporate job with microservices on RMQ and REST APIs. But I always had the desire to experiment with some different tech on the side.<p>After my friend shared his plans of helping facilitate medical conference which was bilingual and some of the audience members might feel missed out due to English not being their first language, I thought it&#x27;s a nice opportunity to create a tool for it. It was an opportunity to work with audio, ASR, accessibility and websockets. It&#x27;s something new for me.<p>Another inspiration was Formula 1 captions which I noticed during their broadcasts, which are generated in real-time. After digging a bit into underlying technology I thought to create something similar for event organisers and AV teams. Not as a replacement for human interpreters, but as a more accessible alternative to quite costly hardware options.<p>That&#x27;s how Stage Captions was built. Audio comes from browser, gets streamed via WSS to backend, goes through processing pipeline and then captions are streamed back to viewers. Those captions can be displayed on venue TVs, opened on attendees devices by scanning QR code or incorporated into a broadcast using OBS for example.<p>Obviously, ASR and MT models are third party and not created by me, but there were still some fun challenges like scaling websockets (it was trickier than just adding one more container on load), audio processing (finding optimal chunking and buffering approach) and obviously stabilization algorithms for partial transcription &#x2F; translation results.<p>As a developer I was happy to experiment with tech stack which is different from my daily work. And as a human it felt right to build something which can make in-person events easier to follow for non-native speakers and deaf&#x2F;hard of hearing attendees.

中文概览

中文标题： Show HN：Stage Captions——把浏览器音频流式转为现场活动字幕

作者是.NET工程师，为一场双语医疗会议做字幕工具。浏览器音频经WSS流到后端，通过处理管线后把字幕流回观众，可显示在场馆电视、扫码在设备上打开，或用OBS并入直播。ASR与MT模型为第三方，难点包括WebSocket扩展、音频分块与缓冲，以及部分转写/翻译结果的稳定化。定位是比昂贵硬件更易获得的选择，而非取代人工口译。

  • 情报分类：开源项目与落地
  • 分类依据：个人开发的实时活动字幕工具发布，属落地项目。
  • 信息来源：Hacker News 新项目
  • 发布时间：2026/9/15 17:41:55