AnySpeechRecognizer lets Swift apps swap speech engines
Swift apps can swap their speech engine
Apple's SpeechAnalyzer API has only one engine behind it: Apple's. A new package called AnySpeechRecognizer keeps that API and lets developers choose the engine.
It comes from Mattt, who wrote the announcement on the Hugging Face blog, alongside a second package, AnyDecisionModel. AnySpeechRecognizer mirrors SpeechAnalyzer and related types, with Apple's Speech framework as the default. For code that uses that shared API, the first change is the import. To use Parakeet, you change how the transcriber is created. Installing assets, running the analyzer, and reading results stay the same.
There are several engines to pick from. Parakeet v3 is NVIDIA's model for 25 European languages. Parakeet Ultra is Moondream's post-trained version of v3, which the post says was more accurate in most of the team's measurements. Parakeet Unified is English-only. Qwen3-ASR is Alibaba's model, covering 30 languages and 22 Chinese dialects. It runs on the GPU and is slower than Parakeet, so the post says it suits files better than live input.
The package also adds speaker diarization, which tracks who speaks when. Both packages can download open models from the Hugging Face Hub and run them on device, and downloads happen only when an app asks for them.
Would you pick a different speech engine for your app?
Video: huggingface.co/blog