發生什麼事
Hugging Face announced two Swift packages, not new models, in its Introducing AnySpeechRecognizer and AnyDecisionModel: Open Models for Speech and Decisions on Apple Platforms post. Both packages can fetch open models from the Hugging Face Hub and run them on-device.
AnySpeechRecognizer builds on the speech-to-text API introduced with macOS 26 and iOS 26. Apple’s engine remains the default; NVIDIA’s Parakeet and Alibaba’s Qwen3-ASR can be selected for on-device use. The package adds speaker diarization, which Apple’s framework does not offer. An app already using the shared API can switch to this package by changing its import. Choosing a different engine also means changing how the transcriber is created.
AnyDecisionModel handles yes-or-no, multiple-choice, and rating questions by using the model’s probabilities for each answer instead of generating text and parsing it. Its six backends include MLX, llama.cpp, and Ollama.
01Sources
- Introducing AnySpeechRecognizer and AnyDecisionModel: Open Models for Speech and Decisions on Apple Platforms: Hugging Face’s official blog, 2026-10-29 (the page gives a date only); package features, engine and backend options, speaker diarization, system requirements, reported results, and test scope