01
Who it’s for
Someone who needs a checked transcript and English translation of long Hindi or Punjabi recordings.
Translation and immigration desks · live
Let three transcripts vote. Check only where they disagree.
For someone who needs a checked transcript and an English translation of long Hindi or Punjabi recordings. They used to listen through the whole recording to check one transcript. I built a tool that has several services transcribe it, combines them word by word, and points a person only at the spots where they disagree.
Runs on my Mac (Apple silicon)Shared with testers by a temporary linkLive · private
Audio → Text → English
transcriber Atranscriber Btranscriber Ctranslation key
sample-call.m4a
made up · two speakers · Hindi0:21Everything here is made up: the clip, the speakers, the Hindi lines and all three “transcribers”. There is no recording; Play makes a soft synthesised tone, only when you press it. The real tool sends real audio to real services. Keys: Space plays, N next step; when checking, J / K, 1–3, Enter.
What it’s for
Four parts, plainly.
02
By hand, until now
They listened through the whole recording to check a transcript.
03
What I built
It removes the silence, has several transcription services transcribe the same audio, and combines them word by word by majority vote, flagging only where they disagree. Then it exports the transcript, subtitles, Word documents and, with a key, an English translation.
04
Where it is
Live and private: it runs on my Mac (Apple silicon), and I can share it with a tester through a temporary link.
Rough edges
What it needs, and where it can’t go.
- Three votersThe vote needs at least three transcription services with API keys to correct anything.
- My Mac is the serverIt cannot run on a normal web host; the Mac has to be the server.
Honest line: voting fixes the mistakes the services make on their own, not the ones they share. That is why every disagreement is shown to a person instead of being quietly settled.