Introducing AnySpeechRecognizer and AnyDecisionModel: Open Models for Speech and Decisions on Apple Platforms
huggingface ·
The article introduces AnySpeechRecognizer and AnyDecisionModel, two new Swift packages designed to bring flexible speech recognition and decision-making capabilities to Apple platforms. AnySpeechRecognizer extends Apple's native SpeechAnalyzer API by allowing developers to choose between different engines, including NVIDIA's Parakeet and Alibaba's Qwen3-ASR, while also adding speaker diarization support via NVIDIA's Streaming Sortformer. These models are converted to Core ML format and hosted on the Hugging Face Hub, enabling local execution with features like live transcription and multi-language support. AnyDecisionModel provides a typed Swift API for zero-shot classification, allowing apps to ask yes-or-no, multiple-choice, or rating-scale questions without generating text. It supports various backends such as MLX, Core AI, llama.cpp, Jev, and Ollama, and can utilize specialized models like Cloudflare's Clef for trained decision heads. The package includes tools for calibrating probabilities and debiasing positional preferences to improve accuracy. Both packages require Swift 6.2 and macOS 26 or iOS 26, leveraging Apple Silicon for optimal performance with local models.
huggingface