Hugging Face Contributor Releases AnySpeechRecognizer and AnyDecisionModel for Apple Platforms
Frontier AI models · AI coding
Author Mattt announced AnySpeechRecognizer and AnyDecisionModel, two Swift packages enabling developers to run open speech and decision models on Apple platforms using Core ML and MLX.
Author Mattt announced two open-source Swift packages: AnySpeechRecognizer and AnyDecisionModel. AnySpeechRecognizer mirrors Apple's SpeechAnalyzer API, allowing developers to swap Apple's speech engine for open models such as NVIDIA's Parakeet variants and Alibaba's Qwen3-ASR running on Core ML, while adding speaker diarization through NVIDIA's Streaming Sortformer. AnyDecisionModel provides typed Swift interfaces for zero-shot decision tasks, reading token probabilities from backends like MLX and llama.cpp without parsing generated free-form text.
Key takeaways
- AnySpeechRecognizer mirrors Apple's SpeechAnalyzer API, allowing developers to switch between Apple's speech framework, NVIDIA Parakeet, and Alibaba Qwen3-ASR on Core ML.
- Speaker diarization is integrated using NVIDIA's Streaming Sortformer on Core ML, enabling tracking of up to four speakers on the same audio timeline.
- AnyDecisionModel extracts token probabilities instead of generating free-form text, returning typed results like Swift enums via backends such as MLX or llama.cpp.
Source：Hugging Face / authors Read the original ↗
Prepared, checked and automatically published by AI; not individually reviewed by a human.