AuK Generate / Edit_Doc

Run instruction TTS, zero-shot TTS, or speech editing. The source/reference and generated target share one 30-second sequence budget.

By DocWorkBox·Created 2 days ago·Updated 2 days ago· 0

◄instructionGenerate speech based on the following description: "a calm, warm female voice". The content to speak is: "Hello, welcome to AuK.".►

◄generation_seconds0.0►

◄use_prompt_enhancertrue►

◄seed42►

◄nfe_steps32►

◄cfg_strength2.0►

◄sway_sampling_coef-1.0►

CategoryAuK_Doc

Inputs (10)

Name	Type	Default	Description
engine	AUK_ENGINE_Doc		—
instruction	STRING	Generate speech based on the following description: "a calm, warm female voice". The content to speak is: "Hello, welcome to AuK.".	Natural-language AuK request. Prompt Enhancer can convert a free-form request into the model instruction.
generation_seconds	FLOAT	0.00–30	Generated target duration in seconds. 0 lets Prompt Enhancer estimate it. When Prompt Enhancer is disabled, enter a value above 0.
use_prompt_enhancer	BOOLEAN	true	Uses the same PE preparation path as the AuK Gradio demo. Credentials are read from server environment variables.
seed	INT	420–9223372036854776000	Seeds both reference VAE encoding and target sampling.
nfe_steps	INT	324–64	Base AuK sampling steps. AuK-Flash requires 4.
cfg_strength	FLOAT	2.00–5	Base AuK classifier-free guidance. AuK-Flash requires 0.
sway_sampling_coef	FLOAT	-1.0-1–1	Base AuK sway sampling coefficient. AuK-Flash requires the default -1 placeholder.
input_audioopt	AUDIO		Reference audio for zero-shot TTS or source audio for editing. Leave unconnected for instruction TTS. Batch size must be 1; channels are averaged to mono.
llm_configopt	AUK_LLM_CONFIG_Doc		Connect AuK OpenAI Settings_Doc or AuK Llama.cpp Adapter_Doc for Prompt Enhancer. Unconnected uses environment variables. Ignored when Prompt Enhancer is disabled.