DictionLabs/sensevoice-small-coreml

SenseVoiceSmall — Core ML (Diction mirror)

The Core ML build that the Diction iOS app downloads for its on-device Chinese and Cantonese dictation model. It is a byte-identical copy of the files the app loads — SenseVoicePreprocessor.mlmodelc, SenseVoiceSmall_int8.mlmodelc, vocab.json, config.json — from FluidInference/sensevoice-small-coreml (revision 74bdbe74b2258f25b5400db72f6cbfbef7c74ded), mirrored so the app downloads from a namespace we control. Load it with FluidAudio's SenseVoiceModels / SenseVoiceManager (precision int8).

License & attribution

SenseVoiceSmall is a model by Alibaba Group (Tongyi Lab, FunAudioLLM): FunAudioLLM/SenseVoiceSmall, part of SenseVoice / FunASR. It is released under the FunASR Model Open Source License Agreement v1.1, Copyright (C) 2023-2028 Alibaba Group; a copy is in LICENSE. As that licence requires, this mirror keeps the model name and the source and author attribution. The Core ML conversion (format only, no retraining; int8 post-training weight quantisation) is by FluidInference. Retain this attribution when redistributing.