speech-recognition

Live speech-to-text transcription on iOS uses AVAudioEngine to capture microphone audio and stream it to the Speech framework, providing continuous partial and final text results. You must manage audio session categories and request microphone permissions before starting real-time dictation.

Pre-recorded audio file transcription uses URL-based recognition flows within the iOS Speech framework to process existing audio files. This offline batch processing method bypasses live AVAudioEngine streaming, returning complete text results for recorded audio assets.

Yes, you can force on-device speech recognition in Swift to reduce latency and avoid server-side limitations. This locale-specific approach uses the Speech framework's built-in support for offline transcription, provided the target language has an available on-device recognizer.

The iOS 26 SpeechAnalyzer module supports async/await for speech-to-text tasks, enabling concurrency-friendly transcription workflows. This modern API replaces older SFSpeechRecognizer callbacks with composable module-based analysis for streaming live microphone input or file recognition.

Speech-to-text recognition fails when required Speech framework and microphone authorization requests are not properly handled before audio capture. You must explicitly request user permissions for both speech recognition and microphone access through AVAudioEngine to prevent recording failures.

Partial speech recognition results provide real-time intermediate text updates during live dictation, while final results represent the completed transcription for a spoken segment. Correctly handling both states in your Swift code ensures responsive and accurate text streaming from AVAudioEngine input.