Neat to see a new implementation, although I'll note that for those looking for a drop-in replacement for the whisper <i>library</i>, I believe that both <a href="https://github.com/guillaumekln/faster-whisper">https://github.com/guillaumekln/faster-whisper</a> and <a href="https://github.com/m-bain/whisperX">https://github.com/m-bain/whisperX</a> are easier (PyTorch-based, doesn't require a web browser), and a lot faster (WhisperX is up to 70X realtime).
Vercel demo doesn't work in Firefox:<p>"TypeError: i[e] is undefined"<p>and "No Model Loaded" error after loading and picking a file and pressing transcribe.