Add speech-rate slider (pitch-preserving time-stretch post-processing, shared across tabs)

Add adjustable cfg_value / inference_timesteps sliders (shared across tabs, defaults = model-card recommendation)

Voice clone: single reference-clip mode only (drop centroid/ECAPA stable mode + speechbrain)

Update for checkpoint step_0006000

Recreate Space (zero-a10g) after region corruption; streaming included

initial commit