SpeechAnalyzer Speech-to-Text

Implement speech-to-text with Apple's new SpeechAnalyzer API (iOS 26+). Powers Notes, Voice Memos, Journal, and Call Summarization.

Overview

SpeechAnalyzer is Apple's next-generation speech-to-text API:

• On-device processing – Private, no server required
• Long-form audio – Optimized for meetings, lectures, conversations
• Distant audio – Works well with speakers across the room
• Volatile results – Real-time approximate results while processing
• Timing metadata – Sync text with audio playback
• Model management – System handles model downloads and updates

When to Use This Skill

Use when you need to:

• ☑ Transcribe live audio (microphone)
• ☑ Transcribe audio files
• ☑ Build Notes-like or Voice Memos-like features
• ☑ Show real-time transcription feedback
• ☑ Sync transcription with audio playback
• ☑ Choose between SpeechAnalyzer and SFSpeechRecognizer

Example Prompts

• "How do I add speech-to-text to my iOS app?"
• "What's the difference between SpeechAnalyzer and SFSpeechRecognizer?"
• "How do I show real-time transcription while recording?"
• "How do I handle volatile vs finalized transcription results?"
• "How do I sync transcript text with audio playback?"
• "Why am I getting insufficientResources from SpeechAnalyzer?"
• "How many transcribers can I run at once?"
• "Why does transcription work on my Mac but fail on a real iPhone?"
• "Why did my audio session break after I added transcription?"
• "How do I transcribe straight from the mic or a video's audio track on iOS 27?"

Key Decision Trees

SpeechAnalyzer vs SFSpeechRecognizer

Need speech-to-text?
├─ iOS 26+ only?
│   └─ Yes → SpeechAnalyzer (preferred)
├─ Need iOS 10-25 support?
│   └─ Yes → SFSpeechRecognizer (or DictationTranscriber)
├─ Long-form audio (meetings, lectures)?
│   └─ Yes → SpeechAnalyzer
└─ Distant audio (across room)?
    └─ Yes → SpeechAnalyzer

Common Use Cases

File Transcription (Simplest)

import AVFoundation
import Speech

func transcribe(fileURL: URL, locale: Locale) async throws -> AttributedString {
    let transcriber = SpeechTranscriber(locale: locale, preset: .transcription)

    async let result = try transcriber.results
        .reduce(AttributedString()) { $0 + $1.text }

    let analyzer = SpeechAnalyzer(modules: [transcriber])

    // analyzeSequence(from:) takes an AVAudioFile — not a URL.
    let file = try AVAudioFile(forReading: fileURL)

    if let lastSample = try await analyzer.analyzeSequence(from: file) {
        try await analyzer.finalizeAndFinish(through: lastSample)
    } else {
        await analyzer.cancelAndFinishNow()
    }

    return try await result
}

SpeechTranscriber's presets are .transcription, .transcriptionWithAlternatives, .timeIndexedTranscriptionWithAlternatives, .progressiveTranscription, .timeIndexedProgressiveTranscription. There is no .offlineTranscription.

Live Transcription Setup

// 1. Configure transcriber with volatile results
let transcriber = SpeechTranscriber(
    locale: Locale.current,
    transcriptionOptions: [],          // required — this init has no defaults
    reportingOptions: [.volatileResults],
    attributeOptions: [.audioTimeRange]
)

// 2. Create analyzer
let analyzer = SpeechAnalyzer(modules: [transcriber])

// 3. Get required audio format
let format = await SpeechAnalyzer.bestAvailableAudioFormat(
    compatibleWith: [transcriber]
)

// 4. Ensure model is available
if let downloader = try await AssetInventory.assetInstallationRequest(
    supporting: [transcriber]
) {
    try await downloader.downloadAndInstall()
}

// 5. Start analyzer
let (stream, builder) = AsyncStream<AnalyzerInput>.makeStream()
try await analyzer.start(inputSequence: stream)

Handle Results

for try await result in transcriber.results {
    if result.isFinal {
        // Finalized - won't change
        finalTranscript += result.text
        volatileTranscript = AttributedString()
    } else {
        // Volatile - will be replaced
        volatileTranscript = result.text
    }
}

Common Pitfalls

• ❌ Forgetting to call finalizeAndFinishThroughEndOfInput() (loses volatile results)
• ❌ Not converting audio to bestAvailableAudioFormat
• ❌ Skipping model availability check before use
• ❌ Not clearing volatile results when finalized arrives
• ❌ Assuming insufficientResources can be caught as catch SFSpeechError.insufficientResources — it can't; that shorthand doesn't compile. Spell the Code type: catch SFSpeechError.Code.insufficientResources
• ❌ Using providerWithSession(...) (iOS 27) when your app owns its audio session — it reconfigures your default AVAudioSession. Use provider(from:in:) and add its captureAudioDataOutput to your own session. (On visionOS that escape hatch does not exist — providerWithSession is the only option there.)
• ❌ Reading AnalyzerInput.buffer (deprecated iOS 27) for duration or format — each access copies the audio. Read bufferDuration / bufferFormat

Simultaneous Analyses

SpeechAnalyzer caps how many backing engines and models it will allocate at once. Apple puts it at roughly two ongoing recognition instances on iOS and visionOS, with currently no limit on macOS — so the same code can pass on a Mac and fail on a real iPhone. Exceeding the cap throws SFSpeechError.Code.insufficientResources.

The cap counts incompatible work: transcribers configured similarly (same locale, same settings) can share backing engines, so making your analyzers alike is the cheap first fix. SpeechAnalyzer.Options.ignoresResourceLimits (iOS 27) opts out of the counting — but it does not raise the hardware ceiling, so you trade a clean, early, catchable error for an unpredictable one later. This dates from iOS 26; it is not new in 27.

Platform Support

Feature	Availability
SpeechTranscriber	iOS 26+, macOS 26+ (not watchOS)
DictationTranscriber	iOS 26+, macOS 26+ (not watchOS, not tvOS)
SpeechAnalyzer	iOS 26+, macOS 26+ (not watchOS)
CaptureInputSequenceProvider / AssetInputSequenceProvider / AnalyzerInputConverter	iOS 27+ (not watchOS)
SFSpeechRecognizer	iOS 10+ (legacy)

Related

• CoreML – deploy custom speech/audio ML models when SpeechAnalyzer doesn't meet your needs
• Foundation Models – generate summaries or titles from transcribed text using Apple Intelligence

WWDC Sessions

• WWDC25-277: Bring advanced speech-to-text with SpeechAnalyzer

Apple Documentation

• Speech Framework