GlmVoiceDecoding

Enums · AiDotNet.TextToSpeech.MultiModal

How GLM-4-Voice’s decoder turns a sequence of speech tokens into audio.

Fields

Field Summary
GlmVoiceDecoding Offline The release’s offline_inference: one flow pass over all the tokens, then HiFT.
GlmVoiceDecoding PaperBlocks The paper’s streaming decoder (§3.2), as the release’s AudioDecoder.stream_inference runs it: blocks of b = 0.8 s (10 tokens, the encoder’s attention block).
GlmVoiceDecoding ReleaseStreaming The release’s web demo (web_demo.py with AudioDecoder.token2wav): blocks of 25, 50, 100, 150 and then 200 tokens.
int value__