Speakers and channels

#Speaker labels

Set diarization: true to label segments with the speaker who said them.

{ "audio": { "upload_id": "upl_…" }, "language": "hi-IN", "diarization": true }

Segments then carry speaker (speaker_0, speaker_1, …) and overlap: true where people talk over each other:

{ "id": 4, "start": 12.4, "end": 15.9, "speaker": "speaker_1", "text": "जी, मैं भेज देता हूँ।" }

Speaker labels work best with two to five speakers. Labels are consistent within one recording, not across recordings.

#Separate channels

Call recordings are usually stereo with the agent on one channel and the customer on the other. Set channel_mode: "separate" to transcribe each channel on its own; every segment then carries channel (0 or 1) and nobody's words are lost when people talk at once.

{ "audio": { "upload_id": "upl_…" }, "language": "ta-IN", "channel_mode": "separate" }

Segments of both channels are returned in time order, and text joins them in that order. Separate mode supports up to two channels and is billed per channel transcribed.

With the default channel_mode: "mix", multi-channel audio is mixed to one channel first (the result carries the channels_mixed warning).

#Which to use