← News

Models & toolsSources: huggingface.co · 29 Oct 2026

New Swift packages bring open speech recognition and decision models to Apple platforms

The AnySpeechRecognizer and AnyDecisionModel packages extend Apple's Swift model APIs with a choice of engines, including NVIDIA, Alibaba, and Cloudflare models running on device.

In plain words

Two new open-source Swift libraries let app developers swap in different AI models for transcribing speech and for answering classification questions on Apple devices. AnySpeechRecognizer keeps Apple's speech-to-text interface but allows engines like NVIDIA's Parakeet, while AnyDecisionModel answers yes-or-no or multiple-choice questions using model probabilities instead of generated text.

What was announced

The authors announced two Swift packages that follow the approach of their AnyLanguageModel library: familiar Apple-style APIs where the underlying model is interchangeable. AnySpeechRecognizer mirrors Apple's SpeechAnalyzer speech-to-text API, introduced in macOS 26 and iOS 26, but adds a choice of recognition engine. It ships with Apple's framework as the default and supports NVIDIA's Parakeet models and Alibaba's Qwen3-ASR running locally on Core ML, plus a speaker diarization module that Apple's API lacks. The team converted the models to Core ML and published them on the Hugging Face Hub, with downloads of roughly 600 MB per Parakeet model and 0.9 to 2.4 GB for Qwen3-ASR.

In the authors' measurements on an M4 Max Mac Studio, Parakeet v3 reached a 1.55% word error rate on LibriSpeech test-clean from files, compared with 2.12% for Apple's transcriber, though all engines they tested scored above 23% on overlapping meeting speech. The second package, AnyDecisionModel, answers yes-or-no, multiple-choice, and rating questions with typed results and probabilities, deriving answers from model probabilities rather than generated text. It supports six backends, including local MLX execution and Cloudflare's Clef-Flash model with a trained decision head. The authors caution that the benchmarks come from one machine and a few datasets, and that they have tested only on macOS 27, not iPhone.

What it does not show

  • The benchmarks were run on a single M4 Max Mac Studio with a few datasets, so results may not generalize to other hardware or audio, as the authors themselves note.
  • Testing covered only macOS 27; no iPhone testing is reported.
  • The source is a blog post by the package authors, so the accuracy and performance claims have not been independently verified.

For security and IT teams

Scope
Both packages require Swift 6.2 or later and macOS 26 or iOS 26 or later; the Parakeet and Sortformer runtimes and the MLX backend require Apple silicon.
Status
AnySpeechRecognizer is versioned from 0.1.0 and AnyDecisionModel from 0.6.0; AnyDecisionModel's CoreAI trait currently cannot be combined with AnySpeechRecognizer due to a swift-huggingface version conflict.

Sources

Source-support review completed 14 Nov 2026. This story was drafted with AI assistance and checked against retained source passages. Automated checking is not independent verification.

Coverage: huggingface.co: selected passages.

Updates and corrections

No corrections so far.

First published 14 Nov 2026

New to the subject? AI from the Ground Up starts from the beginning.