GlmVoiceGeneration
Enums · AiDotNet.TextToSpeech.MultiModal
How GLM-4-Voice lays out its spoken reply (Zeng et al. 2024b §3.3).
Fields
| Field | Summary |
|---|---|
GlmVoiceGeneration StreamingThoughts |
The paper’s “streaming thoughts”: after <|assistant|>streaming_transcription the reply alternates 13 text tokens and 26 speech tokens, so speech starts before the text is complete. |
GlmVoiceGeneration TextFirst |
The decoupled two-step generation (§3.3, with the published template of Zeng et al. |
int value__ |