speech-to-text

Use SFSpeechRecognizer with AVAudioEngine for deployment targets below visionOS 26, or SpeechAnalyzer with SpeechTranscriber on visionOS 26+. Add the microphone and speech recognition usage strings to Info.plist, request both permissions, then stream partial transcription results into your text field.

Choose by deployment target: SFSpeechRecognizer works on visionOS 1.0+ and is the broadly compatible default, while SpeechAnalyzer with SpeechTranscriber is faster and better for long-form transcription but requires visionOS 26+. Gate the modern API with #available(visionOS 26, *).

The crash happens when you access engine.inputNode before activating AVAudioSession, leaving the input node without a valid hardware format. Configure and activate the audio session first, and pass nil as the tap format instead of outputFormat(forBus:).

You need NSMicrophoneUsageDescription and NSSpeechRecognitionUsageDescription. Accessing the microphone or speech APIs without these strings crashes the app, and you must also request both authorizations at runtime before starting capture.

No, SpeechTranscriber and SpeechAnalyzer are only available on visionOS 26 and later. For earlier targets, use SFSpeechRecognizer as the fallback branch, and check SpeechTranscriber.supportedLocales plus install model assets before first use.