Murmur

Speech model (downloaded on first launch, not bundled with the app):

  Parakeet TDT 0.6B v3 by NVIDIA
  https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3
  Licensed under the Creative Commons Attribution 4.0 International License (CC BY 4.0):
  https://creativecommons.org/licenses/by/4.0/

  Core ML conversion by FluidInference
  https://huggingface.co/FluidInference/parakeet-tdt-0.6b-v3-coreml
  (commit 7dd20fe6b1797d35f5e3307e8b1732d9a178edfe), CC BY 4.0.

  The model files are used unmodified.

Cleanup model (downloaded on first launch from Murmur's own server, murmur-cdn.sathuashrith.com, not bundled):

  A small language model that turns the transcript into the text to paste (fillers, repeats, self-corrections,
  lists, numbers, the user's dictionary words, the app's writing style). Fine-tuned from
  Spokenly Cleanup 1 by Spokenly (Apache License 2.0)       https://huggingface.co/spokenly/cleanup-1
  itself fine-tuned from Qwen3-0.6B-Base by Qwen (Apache License 2.0)   https://huggingface.co/Qwen/Qwen3-0.6B-Base
  then converted to Core ML with int8 weights. Training, evaluation and export: ml/llm/README.md.
  The model is a modified version of that work (fine-tuned, converted and quantised). The full text of the
  Apache License 2.0 ships with the app as LICENSE-Apache-2.0.txt (Settings → About) and is in the source tree at
  assets/licenses/Apache-2.0.txt. https://www.apache.org/licenses/LICENSE-2.0

  The fine-tuned weights are licensed under the Creative Commons Attribution-ShareAlike 4.0 International
  License (CC BY-SA 4.0): https://creativecommons.org/licenses/by-sa/4.0/
  They were trained on data adapted (answers restyled, prompts reformatted) from:
  - dictation-to-text, ductai199x, CC BY-SA 4.0 (hence the ShareAlike licence of the weights)
    https://huggingface.co/datasets/ductai199x/dictation-to-text
  - Wikipedia (English, 20231101 dump), Wikipedia contributors, CC BY-SA 4.0, sentences spoken by Kokoro-82M
    (hexgrad, Apache License 2.0) and transcribed by Parakeet; the audio was never kept
    https://dumps.wikimedia.org

app/Engine/ParakeetTranscriber.swift drives these models. It contains no code copied from FluidAudio
(https://github.com/FluidInference/FluidAudio, Apache License 2.0), which was used as a reference for the
models' inputs and outputs.

Licence texts: the Apache License 2.0 is bundled in full (LICENSE-Apache-2.0.txt). The CC BY 4.0 and
CC BY-SA 4.0 legal code is linked above, as both licences allow (section 3(a)(1)(C)).