# Fluid Speech Pro · Clean

Complete Fluid Speech Pro for Glade, including FrameCleaner.
This is the clean transcription mode; metadata.json declares its identity.

30-second contiguous quiet-boundary chunks, W8 encoder projections, FP16 graph
tensors, six four-layer encoder stages and an ANE TDT decoder.

The trained clean adapter is merged offline; no postprocessing model is needed.

Source: https://huggingface.co/FluidVoice-Labs/fluid-speech-pro
Source revision: 46027a3bffc5817ce8f9b29add0a062610134a90.

CC-BY-4.0; see LICENSE and NOTICE.

Fluid Speech Pro by FluidVoice. Built on NVIDIA Parakeet TDT 0.6B v3 and
Moondream Parakeet Ultra.