HappySRT

HappySRT's whole design premise is that transcription shouldn't require handing your recording to a third party. The entire pipeline — FFmpeg decoding, on-device speech recognition, translation, and summarization — runs inside your browser tab. You can verify this by reading the source, since it's MIT-licensed. For anyone handling confidential interviews, medical conversations, legal recordings, or unpublished research, that's a substantive advantage over cloud services that store your uploads. The workflow is deliberately narrow: load a file, wait for the engine to initialize, pick transcription / translation / summarization, export as text or SRT. That covers a lot of real-world jobs — podcast quote extraction, interview transcripts, subtitle files for short films and vlogs, lecture summaries, conference talk captioning. Where it falls short is equally clear. Transcription quality and speed depend entirely on your own hardware; a long recording on a modest laptop will take real time, and there's no cloud fallback when your machine struggles. There is no speaker diarization, so multi-person interviews come out as one continuous block of text. There is no real-time captioning, no shared workspace, and no timestamp editing or segment merging inside the tool once you have your SRT. The interface is minimal by design. That makes HappySRT a specialist tool rather than a general-purpose transcription platform. If your work is single-user, privacy-critical, and doesn't require labelling who said what, it's a clean fit and costs nothing. If your work involves team review, multi-speaker attribution, or live captioning, you're better served by Otter.ai or Descript — and you should weigh the fact that both route your media through their infrastructure.