As someone who built my own internal self host infrastructure for transcriptions with speaker diarisation, my own app for myself, my own 3-2-1 archive solution, using runpod and modal alongside my own VPS and at he hardware…
I’ll say i’m interested. However I also know the big challenge with this is the tradeoff of time. You can run WhisperX or Parakeet locally, but it takes longer than using proper hardware. If you don’t need it in minutes after the meeting /conference/ whatever, then it can work. Shortcutting around it by model reductions, just makes it less accurate which has it’s own consequences.
As someone who built my own internal self host infrastructure for transcriptions with speaker diarisation, my own app for myself, my own 3-2-1 archive solution, using runpod and modal alongside my own VPS and at he hardware…
I’ll say i’m interested. However I also know the big challenge with this is the tradeoff of time. You can run WhisperX or Parakeet locally, but it takes longer than using proper hardware. If you don’t need it in minutes after the meeting /conference/ whatever, then it can work. Shortcutting around it by model reductions, just makes it less accurate which has it’s own consequences.
Overall this field is ripe for self-hosting.