Downloads · 30 days
18
28% of all-time downloads
parakeet-tdt-0.6b-v3-litert-dynamic-int8
raspbfox/parakeet-tdt-0.6b-v3-litert-dynamic-int8
parakeet-tdt-0.6b-v3-litert-dynamic-int8 is a automatic speech recognition model from raspbfox. Use it when you need speech turned into text. It is set up for litert. The card lists the license as cc-by-4.0.
Experimental 10-second fixed-window LiteRT conversion of nvidia/parakeet-tdt-0.6b-v3.
All-time downloads
65
Public
Repo size
617 MB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.tflite617 MB · 100%
At a glance
- Task
- Automatic Speech Recognition
- Library
- litert
- License
- cc-by-4.0
- Access
- Public
- Created
- Jul 13, 2026
- Updated
- Jul 13, 2026
- SHA
- 01b0c122
Base models
Model card
From the Hugging Face model README
Parakeet-TDT 0.6B v3 — split LiteRT dynamic INT8
Experimental 10-second fixed-window LiteRT conversion of
nvidia/parakeet-tdt-0.6b-v3.
Generated with Google litert-torch commit
4502654d6afe38b495d2efe8dcd37e057d0c5e7f and the
dynamic_wi8_afp32 quantization recipe.
encode_quantized.tflite:[1, 128, 1001]→[1, 1024, 126]decode_quantized.tflite: encoder output, one float32 token ID, and two[2, 1, 640]states →[1, 126, 1, 8198]logits and next states
The token ID is exposed as float32 and cast to int32 inside the decoder graph so
Flutter's LiteRT CompiledModel API can use uniform float buffers. Split files
allow independent NPU/CPU placement.
This is an exact-parity development artifact, not the final calibrated A8W8 NeuroPilot release. On the NVIDIA sample WAV it produced the exact same 44 token IDs and transcript as NeMo.
- Task
- Automatic Speech Recognition
- Library
- litert
- License
- cc-by-4.0
- Languages
- en, de, es, fr, it, pt
- Created
- Jul 13, 2026
- Updated
- Jul 13, 2026