GoogleDialogueSpeakerMode

Namespace Tracon · Assembly Tracon.Google.dll

How a Gemini dialogue request names its speakers.

public enum GoogleDialogueSpeakerMode

Fields

PartMetadata = 0

Each turn is a text part with its speaker in the part’s speech metadata. The format of the gemini-3.8 models.

TextLabels = 1

The turns are lines of one text, each starting with its speaker label (Speaker1: …). The format of the gemini-2.5 and gemini-3.1 preview models.

A turn whose text has a line that starts with a speaker label would change the voice, so Tracon refuses it before the request is sent.

Remarks

The two formats exist because the model generations disagree (measured 2026-10-25): the gemini-3.8 models want the speaker on every text part and refuse a speaker label in the text; gemini-2.5-flash-preview-tts and gemini-3.1-flash-tts-preview want the label in the text and refuse the part metadata.