Comment by MiroF

Why can't they just have full on-device transcription models? It's lossy, but extremely compressed.