← Back
Hugging Face Blog

Introducing AnySpeechRecognizer and AnyDecisionModel: Open Models for Speech and Decisions on Apple Platforms

•11 min read•
#llm
Level:Advanced
For:AI Engineers
✦TL;DR

Introducing AnySpeechRecognizer and AnyDecisionModel adds open‑source speech‑to‑text and decision‑making engines to Apple’s SpeechAnalyzer API, letting developers swap Apple’s default engine for NVIDIA’s Parakeet (v3, Ultra, Unified) or Alibaba’s Qwen3‑ASR and run decision models via MLX, Core AI, llama.cpp, Jev, or Ollama. Parakeet v3 (0.6B) supports 25 European languages with automatic language detection, while Qwen3‑ASR offers 30 languages plus 22 Chinese dialects in 0.6B/1.7B sizes but runs slower on GPU and lacks volatile results. Both packages pull models from Hugging Face and run on‑device via Core ML, simplifying integration and reducing latency and cost for speech and decision workloads. Engineers can now replace a single import to switch engines and obtain typed probabilistic decisions without LLM text generation.

⚡ Key Takeaways

  • Parakeet v3 is NVIDIA’s 0.6B model that recognizes 25 European languages and auto‑detects locale.
  • AnySpeechRecognizer mirrors SpeechAnalyzer, SpeechTranscriber, DictationTranscriber, SpeechDetector, and AssetInventory, allowing engine swaps by changing the transcriber creation.
  • Qwen3‑ASR provides 0.6B and 1.7B models covering 30 languages and 22 Chinese dialects, runs on GPU, and is slower than Parakeet, making it better for file transcription.
  • AnyDecisionModel derives yes/no, multiple‑choice, and rating‑scale answers from model probabilities using MLX, Core AI, llama.cpp, Jev, or Ollama backends.
  • Both packages download open models from the Hugging Face Hub and execute them on‑device via Core ML.
💡 Why It Matters

Engineers can now deliver high‑accuracy, low‑latency speech transcription and probabilistic decision outputs on iOS/macOS without relying on cloud services, cutting operational costs and improving privacy.

✅ Practical Steps

  1. Import AnySpeechRecognizer and replace the default transcriber creation with `ParakeetTranscriber()` (or another engine) while keeping the same SpeechAnalyzer workflow.
  2. Use AssetInventory to pre‑download the desired Parakeet or Qwen3‑ASR model from Hugging Face before running the analyzer.
  3. Integrate AnyDecisionModel by selecting a backend (e.g., MLX) and call its `predict(question:)` method to obtain typed answers and probability scores.

Want the full story? Read the original article.

Read on Hugging Face Blog ↗

More like this

Censored vs. Truncated Regression in Python, Explained

Towards Data Science•#llm

How Google Duplex evolved to also call businesses for you in Gemini

Google Research•#agents

Q&A: Enabling more human-like interactions and understanding in multimodal foundation models

MIT News AI•#llm

Welcome Inkling 2 Mini, Cutting-edge Performance for your Local Setup

Hugging Face Blog•#llm

EXPLORE AI NEWS

Daily hand-picked stories on LLMs, RAG, agents and production AI — curated for engineers who ship.

BROWSE NEWS

GET THE WEEKLY DIGEST

Join engineers getting the Monday signal-over-noise AI breakdown. No spam, unsubscribe anytime.

LEARN AI ENGINEERING

Curated courses, research papers, repos and tutorials built for engineers leveling up in AI.

START LEARNING