Medscribe

Offline medical dictation by Aonz

Medscribe models

Speech-to-text models and the language model used by Medscribe, an offline desktop dictation app for radiology and clinical reporting (Windows, macOS, Linux), built by Aonz. The app downloads the files it needs from this repo on first use.

Files

File What it is Size Licence
medasr-Q8_0.gguf MedASR, 8-bit. Default and recommended. 122 MB HAI-DEF
medasr-F16.gguf MedASR, 16-bit 202 MB HAI-DEF
medasr-F32.gguf MedASR, full precision 402 MB HAI-DEF
parakeet-unified-en-0.6b-Q8_0.gguf Parakeet English 0.6B, 8-bit (general English, streaming) 697 MB CC BY 4.0
lm_6.kenlm MedASR's 6-gram KenLM language model for beam search 672 MB HAI-DEF

The .gguf files run with transcribe.cpp. lm_6.kenlm is used by Medscribe's MedASR decoder: CTC prefix beam search with KenLM shallow fusion (beam 8, alpha 0.5, beta 1.5), the same setup Google describes for MedASR. Without it, MedASR falls back to greedy decoding.

SHA-256

5a391a2154416b96241b829ac2c7eed8b64d197f00372512c9bbf63f4849c978  medasr-Q8_0.gguf
70e51cf5da1d3ea0838cf6d94a19f3dc1fdf28b24d8665589f43c50eaf429580  medasr-F16.gguf
4b1b5b3a3092b96299624788b4eb28d04f7c66c447f0953c0216285c518f2ed5  medasr-F32.gguf
4b50b6dd862bf6e346929aaf4f5eaacec003bfa3f56462d6c874b41ef2f38795  parakeet-unified-en-0.6b-Q8_0.gguf
e559df5485675c47f3191f81dd6314160dc4e743f2f6a09dd455d5ac97e928f2  lm_6.kenlm

Sources

Licences

  • MedASR files and lm_6.kenlm are HAI-DEF and are provided under and subject to the Health AI Developer Foundations Terms of Use (copy in LICENSE). By using them you agree to those terms, including their use restrictions. HAI-DEF is not intended for clinical use without appropriate validation and any required regulatory authorization.
  • Parakeet is by NVIDIA, licensed under CC BY 4.0. The GGUF conversion repackages the weights at 8-bit precision; it was not retrained.

See NOTICE for attribution and modification details.

About

Medscribe is made by Aonz. The Medscribe name and logo (assets/) belong to Aonz and are not covered by the model licences above.

Downloads last month
50
GGUF
Model size
0.1B params
Architecture
medasr
Hardware compatibility
Log In to add your hardware

8-bit

16-bit

32-bit

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

Model tree for SavioAonz/medscribe-models

Base model

google/medasr
Quantized
(9)
this model