Run Argmax's WhisperKit on Apple Silicon: Swift package setup, the whisperkit-cli, CoreML model downloads, and fixes for compile-time and first-run stalls.
brew install whisperkit-cli then whisperkit-cli transcribe --audio-path file.m4a. First run downloads + specializes a CoreML model.
Add https://github.com/argmaxinc/WhisperKit in Xcode; iOS 16+/macOS 13+ targets.
Models pull from the Hugging Face argmaxinc repo automatically; --model large-v3 selects size.
Same Whisper-class accuracy, no install, no model downloads, no GPU questions. Speaker labels, word timestamps and every export (TXT, SRT, VTT, DOCX) included. Free instant preview; credits from $2 and they never expire.
CoreML “model specialization” compiles the model for your exact chip on first load — up to several minutes for large-v3. It's cached afterward; don't kill it.
WhisperKit targets the Apple Neural Engine — Intel Macs aren't the audience. Use whisper.cpp or a hosted engine there.
Correct — WhisperKit transcribes only. Pair it with a diarization service, or use a hosted pipeline that returns labeled speakers.
Ship the tiny/base model in-app and download larger models on first launch — Apple's cellular download cap applies to the bundle, not post-install downloads.
More on WhisperKit: the full WhisperKit page · WhisperKit alternatives · all transcription tools