Speech Recognition

Speech Recognition

Convert live speech and prerecorded audio to text. Targets modern on-device SpeechAnalyzer (iOS 26+) and legacy SFSpeechRecognizer (iOS 10+).

Contents

• Permissions & Audio Session
• Modern SpeechAnalyzer (iOS 26+)
• Legacy SFSpeechRecognizer
• On-Device Model Assets
• Common Mistakes
• Review Checklist
• References

Permissions & Audio Session

Add NSSpeechRecognitionUsageDescription and NSMicrophoneUsageDescription to Info.plist. Request speech and microphone authorization before starting recording:

import Speech
import AVFAudio

func requestPermissions() async -> Bool {
    let speechAuthorized = await withCheckedContinuation { continuation in
        SFSpeechRecognizer.requestAuthorization { status in
            continuation.resume(returning: status == .authorized)
        }
    }
    guard speechAuthorized else { return false }
    return await AVAudioApplication.requestRecordPermission()
}

Modern SpeechAnalyzer (iOS 26+)

SpeechAnalyzer delivers modular, asynchronous streaming transcription using SpeechTranscriber:

import Speech

final class LiveTranscriber {
    private var analyzer: SpeechAnalyzer?
    private var transcriber: SpeechTranscriber?

    func startTranscribing(format: AVAudioFormat) async throws {
        let transcriber = SpeechTranscriber(locale: Locale(identifier: "en-US"), preset: .transcription)
        let analyzer = SpeechAnalyzer(modules: [transcriber])
        self.transcriber = transcriber
        self.analyzer = analyzer

        Task {
            for try await result in transcriber.results {
                let text = result.text
                if result.isFinal {
                    print("Committed: \(text)")
                } else {
                    print("Partial: \(text)")
                }
            }
        }

        try await analyzer.start(inputFormat: format)
    }

    func appendAudio(buffer: AVAudioPCMBuffer) {
        analyzer?.append(buffer)
    }

    func stop() async throws {
        try await analyzer?.finish()
    }
}

Legacy SFSpeechRecognizer

For audio file transcription and backward compatibility:

func transcribeFile(url: URL) async throws -> String {
    guard let recognizer = SFSpeechRecognizer(locale: Locale(identifier: "en-US")),
          recognizer.isAvailable else {
        throw SpeechError.unavailable
    }

    let request = SFSpeechURLRecognitionRequest(url: url)
    request.requiresOnDeviceRecognition = true

    return try await withCheckedThrowingContinuation { continuation in
        recognizer.recognitionTask(with: request) { result, error in
            if let error {
                continuation.resume(throwing: error)
            } else if let result, result.isFinal {
                continuation.resume(returning: result.bestTranscription.formattedString)
            }
        }
    }
}

On-Device Model Assets

Check asset installation or download language packs before recognition:

let status = await AssetInventory.status(for: .transcription, locale: Locale(identifier: "en-US"))
if status != .installed {
    try await AssetInventory.download(for: .transcription, locale: Locale(identifier: "en-US"))
}

Common Mistakes

• Missing microphone usage description: Omitting NSMicrophoneUsageDescription crashes immediately when accessing AVAudioEngine.
• Conflating partial and final results: Partial results update frequently; only commit text into data models when result.isFinal is true.
• Starting multiple concurrent recognition tasks: Always finish or cancel the previous task before launching a new recognition stream.
• Ignoring on-device fallback failures: If requiresOnDeviceRecognition is true, ensure model assets are installed via AssetInventory.
• Not stopping audio engine alongside recognition: Failing to halt AVAudioEngine keeps the microphone indicator active in the status bar.

Review Checklist

• NSSpeechRecognitionUsageDescription and NSMicrophoneUsageDescription present in Info.plist
• Speech recognition authorization requested and confirmed authorized
• Audio engine input nodes and tap callbacks cleaned up upon completion
• SpeechAnalyzer (iOS 26+) used for modern streaming pipelines
• Transient partial results visually distinguished from finalized transcripts

References

• SpeechAnalyzer pipelines, volatile results, and audio buffering: references/speechanalyzer-patterns.md
• Speech framework
• SpeechAnalyzer
• SpeechTranscriber
• SpeechTranscriber.Preset
• DictationTranscriber
• SpeechDetector
• SFSpeechRecognizer
• SFSpeechAudioBufferRecognitionRequest
• SFSpeechURLRecognitionRequest
• SFSpeechRecognitionResult
• SFSpeechRecognitionRequest
• AssetInventory
• Asking Permission to Use Speech Recognition
• Recognizing Speech in Live Audio
• Bring advanced speech-to-text to your app with SpeechAnalyzer