markstrom/pianissimo-sv-coreml

Klang Pianissimo — CoreML

A CoreML conversion of Klang Pianissimo by Klang AI AB, a Swedish speech recognition model (FastConformer encoder + TDT decoder, fine-tuned from NVIDIA Parakeet TDT 0.6B v3), for on-device use on Apple Silicon through the Apple Neural Engine.

All credit for the model goes to Klang AI AB. This repository only changes the file format; the weights were not retrained. Licensed under CC BY 4.0, like the original.

Files

File	Purpose
Preprocessor.mlpackage	16 kHz mono audio → 128-band log-mel features
Encoder.mlpackage	FastConformer encoder, int8 weights
Decoder.mlpackage	TDT prediction network
JointDecisionv3.mlpackage	Joint network + decision head
parakeet_vocab.json	Token vocabulary
manifest.json	Version, sizes and checksums

Compile the packages on the device that uses them (MLModel.compileModel(at:)), then load the directory with FluidAudio's AsrModels.loadLocal(from:version: .v3).

See LICENSE-and-attribution.txt for the exact source revision and the list of changes.