Fluid Speech Pro (GGUF, Q8_0)

Fluid Speech Pro by FluidVoice, converted for transcribe.cpp by TalkMate.

Fluid Speech Pro adds one block between Parakeet's subsampling and its first Conformer layer (enc.pre_block.*, announced by stt.parakeet.encoder.pre_block_kernel). It needs a transcribe.cpp built with TalkMate's patch for that block; an unpatched build silently runs without it and transcribes worse.

On accented English (EdAcc, 600 clips): 10.2% WER, against 10.1% for the original .nemo.

Licensed CC BY 4.0. Built on NVIDIA Parakeet TDT 0.6B v3 and Moondream Parakeet Ultra, both CC BY 4.0.

Downloads last month
26
GGUF
Model size
0.6B params
Architecture
parakeet
Hardware compatibility
Log In to add your hardware

8-bit

Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for TalkMate/fluid-speech-pro-gguf

Quantized
(1)
this model