Key points
- AnySpeechRecognizer allows swapping Apple's default engine for NVIDIA or Alibaba models.
- Parakeet v3 supports twenty-five European languages with automatic detection.
- Qwen3-ASR covers thirty languages and twenty-two Chinese dialects.
Hugging Face released AnySpeechRecognizer and AnyDecisionModel to extend local machine learning capabilities on Apple platforms.confirmed These Swift packages provide familiar APIs for speech recognition and decision tasks, allowing developers to choose specific open models instead of relying solely on proprietary engines.
Evidence: confirmed
Introducing AnySpeechRecognizer and AnyDecisionModel: Open Models for Speech and Decisions on Apple Platformsaifleet-huggingface
The evidence
The AnySpeechRecognizer package implements Apple’s SpeechAnalyzer API but permits engine selection. It supports NVIDIA’s Parakeet and Alibaba’s Qwen3-ASR running locally via Core ML. The package also adds speaker diarization to track who speaks when, using models converted to Core ML and published on the Hugging Face Hub.confirmed
Evidence: confirmed
Introducing AnySpeechRecognizer and AnyDecisionModel: Open Models for Speech and Decisions on Apple Platformsaifleet-huggingface
The decision
Developers can switch engines by changing how they create the transcriber, keeping installation and result-reading steps identical. Apple’s Speech framework remains the default. This approach mirrors the strategy used for AnyLanguageModel, aiming to provide faster, less expensive alternatives to general-purpose language models for specific tasks like routing customer support calls.
What the evidence means
Parakeet v3 is a polyglot model recognizing twenty-five European languages without requiring locale input. Parakeet Ultra, a post-trained variant, showed higher accuracy in Hugging Face’s measurements. Parakeet Unified is an English-only option. All three run on a custom Core ML runtime supporting live input, word timing, and confidence scores.
The evidence for research
The research tests the feasibility of replacing Apple’s proprietary speech engine with open-weight models within a standardized Swift API. Reported results indicate Parakeet models handle live input with volatile results and word timing, while Qwen3-ASR provides single final results per segment.reported Limitations include Qwen3-ASR’s slower GPU performance, lack of word timing or confidence scores, and its suitability for file transcription rather than live streaming.
Evidence: reported
Introducing AnySpeechRecognizer and AnyDecisionModel: Open Models for Speech and Decisions on Apple Platformsaifleet-huggingface
Why it matters
- Evaluators can benchmark open models against Apple’s default engine using the provided Core ML implementations.
- The inclusion of speaker diarization allows testing of multi-speaker tracking capabilities in local speech pipelines.
- Qwen3-ASR’s support for Chinese dialects expands test coverage beyond the twenty-five European languages covered by Parakeet.
- AnyDecisionModel enables evaluation of typed, probability-based answers for yes-or-no and rating-scale tasks without text generation.
Topics: Hugging Face, Swift, speech recognition, open models, Apple
Updated
Ask the article
Local keyword search of this article; no generated answers.Which of these is confirmed?
Check the answer
Hugging Face released AnySpeechRecognizer and AnyDecisionModel to extend local machine learning capabilities on Apple platforms.
The matched primary source and announcement wording support the confirmed label.
Did this article give you the evidence you needed?
Your choice stays on this device. No public tally.Questions readers ask
Which languages does Parakeet v3 support?
Parakeet v3 recognizes twenty-five European languages and automatically detects which one is being heard.
Can Qwen3-ASR handle live audio input?
Qwen3-ASR is slower and better suited for transcribing files, as it lacks support for volatile results or live input streaming.
How does AnyDecisionModel generate answers?
It derives answers from model probabilities for yes-or-no, multiple-choice, and rating-scale questions, rather than generating text.
Connected stories
Reading receipt
- AnySpeechRecognizer allows swapping Apple's default engine for NVIDIA or Alibaba models.
- Parakeet v3 supports twenty-five European languages with automatic detection.
- Qwen3-ASR covers thirty languages and twenty-two Chinese dialects.