One POST uploads the whole file; the API splits it into ~384 KB objects in S3 and notifies Centrifugo per chunk so workers can transcribe cumulatively. WAV/PCM-friendly; arbitrary MP3 byte splits may fail until enough data arrives.