DictionLabs / parakeet-tdt-0.6b-v3-coreml

parakeet-tdt-0.6b-v3-coreml

NVIDIA Parakeet in CoreML format, for on-device speech-to-text on Apple hardware. Mirrored by Diction Labs .

This is a mirror, not our work

The model files are an unmodified copy of FluidInference/parakeet-tdt-0.6b-v3-coreml at revision aed02740059203c4a87495924f685de3722ae9ce , byte for byte. Credit belongs to two parties:

• NVIDIA trained the original Parakeet model.
• FluidInference converted it to CoreML, which is the artifact reproduced here.

The only difference from the source repository is this README. Everything else is identical and can be verified by comparing file names and LFS object ids against the revision above.

Licensed CC-BY-4.0, the same as both upstreams. We add no restrictions.

Why mirror it

The Diction app downloads this at runtime through FluidAudio , whose API resolves a repository path but no revision, so it always follows the default branch. Pointing that at a namespace we do not control means any upstream push reaches every user immediately, with no review and no rollback. Mirroring is the only mechanism that gives us a pin: a new upstream release reaches nobody until we have tested it and promoted it deliberately.

Use with Diction

Self-hosting the app is not required to use this model. Point FluidAudio at it with ModelRegistry.repoOverrides :

ModelRegistry.repoOverrides = ["FluidInference/parakeet-tdt-0.6b-v3-coreml": "DictionLabs/parakeet-tdt-0.6b-v3-coreml"]