Batch

Whole-file compatibility: one POST carrying a WAV body.

Whole-file compatibility

POST /v1/transcriptions?mode=mt_asr&expected_num_speakers=auto accepts a WAV body. It creates a normal MT session, feeds 320 ms chunks through the same scheduler, flushes, and collects its events; it is not a second inference implementation.

Target-speaker files

TS whole-file compatibility is deliberately not exposed: large enrollment headers are fragile and privacy-sensitive. Feed the file and enrollment incrementally through /v1/stream instead.