janfang/pianissimo-sv-coreml

Pianissimo-sv – Core ML

This is a copy of the Core ML conversion of KlangAI/pianissimo-sv, Klang AI's Swedish speech-to-text model. Ordfång, a small dictation app by Jan Fång, downloads its model from here, so that the app keeps working if another repository moves or changes.

All credit for the model goes to Klang AI. Nothing in this repository was trained or changed by the owner of this repository.

Attribution

Part	Author	Licence
Pianissimo, the Swedish model	Klang AI	CC BY 4.0
Parakeet TDT 0.6B v3, the base model	NVIDIA	CC BY 4.0
Core ML conversion	magnushjelm	CC BY 4.0

Changes from the original

The files are identical to magnushjelm/pianissimo-sv-coreml at revision ff8af835072b77a880991ce2d5402d249f923886. In that conversion the weights were converted to Core ML and the encoder weights were compressed with 8-bit k-means palettization. The preprocessor and the vocabulary are the unchanged Parakeet v3 files.

Measured

Swedish FLEURS test set, 759 clips, MacBook Air M3: word error rate 6.69 %, 0.08 s per clip. Klang AI reports 6.51 % for the original model.

Use

Load the directory with FluidAudio: AsrModels.loadLocal(from: directory, version: .v3).

Licence

CC BY 4.0