The new packages enable Apple platform developers to execute specialized speech recognition and structured decision-making workloads locally or via external hosts. AnySpeechRecognizer mirrors Apple's SpeechAnalyzer framework to offer drop-in engine replacements alongside speaker diarization capabilities. Meanwhile, AnyDecisionModel bypasses traditional text generation by deriving typed answers directly from underlying model probabilities. The software supports diverse local runtimes such as Core ML, MLX, Core AI, and llama.cpp.
Key points
Developers can execute open speech and decision models on-device using Swift APIs and Hug…: Developers can execute open speech and decision models on-device using Swift APIs and Hugging Face Hub downloads.
Purpose-built models offer faster, more accurate, and cheaper alternatives to general-pur…: Purpose-built models offer faster, more accurate, and cheaper alternatives to general-purpose language models for specialized tasks.
The architectures support flexible backends, ranging from local frameworks like Core ML a…: The architectures support flexible backends, ranging from local frameworks like Core ML and MLX to external hosting via Ollama or Jev.
Sources checked by our analysis pipeline: 1
Language models can do a lot. They write, summarize, translate, and call tools. But they're only a small part of what machine learning can do.
Say a customer calls your support line. First, you want to know what they said (speech recognition). Then, you want to decide where to send them (decision). You could hand both to a language model, but there are models built for each that are better, faster, and less expensive.
In our AnyLanguageModel 1.0 recap, we looked at a year of local and hosted models in Swift. With AnyLanguageModel and swift-huggingface now on their 1.x releases, we're extending the same approach to speech and decisions: familiar Swift APIs with a choice of model.
Today we're announcing two packages:
- AnySpeechRecognizer implements Apple's new speech-to-text API with a choice of recognition engine, including NVIDIA's Parakeet and Alibaba's Qwen3-ASR running locally on Core ML, and adds speaker diarization to track who speaks when.
- AnyDecisionModel answers yes-or-no, multiple-choice, and rating-scale questions with typed results and probabilities. It derives each answer from the model's probabilities instead of generating text,…
Nvidia CEO Jensen Huang has reportedly become entangled in a Chinese spy investigation. The development arrives as artificial intelligence stocks experience a downturn. Details regarding the nature of the probe remain limited.
The strategic deal reunites two 2016 Y Combinator alumni to accelerate modular development in real-time voice and video applications. By absorbing the infrastructure specialist behind Pipecat, the newly minted unicorn is positioning itself as a foundational backbone for agentic workflows across the broader ecosystem. This integration will lower engineering barriers for enterprises deploying interactive conversational agents at scale.
Google's latest multimodal model secured the peak position by leading across four distinct benchmarks focused on complex reasoning, science, and broad knowledge. Accessible initially to select users, the system integrates a 1-million-token context window capable of processing speech, text, video, and image inputs. This milestone reflects significant architectural progress that directly challenges leading competitors in high-level analytical performance.