Modules:
| Name | Description |
|---|---|
duration | Text duration estimation for TTS generation. |
fused_qkv_rope | Fused attention prologue for models with a packed QKV projection. |
omnivoice | OmniVoice model for vLLM-Omni two-stage TTS pipeline. |
omnivoice_decoder | OmniVoice Decoder (Stage 1) - Audio token to waveform conversion. |
omnivoice_generator | OmniVoice Generator (Stage 0) - Iterative unmasking with Qwen3 backbone. |
pipeline | OmniVoice pipeline topology (frozen). |