For what iPhone and iPad models under iOS 26 SpeechTranscriber.isAvailable is true

I am trying to use SpeechTranscriber from Speech framework. Is it possible to use it on Simulator of iOS 26 (Mac OS Tahoe)? Function "supportedLocales" returns an empty array.

Hi everyone 👋 I’m building an iOS app in Swift where I want to do the following: Record the user’s voice Transcribe the spoken sentence (speech-to-text) Auto-detect the spoken language Translate it to another language selected by the user (e.g., English → Spanish or Hindi → English) Speak back (text-to-speech) the translated text on the same device Is this possible to record via phone mic and play the transcribe voice into headphone's audio?

Hello, I am wondering if it is possible to have audio from my AirPods be sent to my speech to text service and at the same time have the built in mic audio input be sent to recording a video? I ask because I want my users to be able to say "CAPTURE" and I start recording a video (with audio from the built in mic) and then when the user says "STOP" I stop the recording.

I've tried SpeechTranscriber with a lot of my devices (from iPhone 12 series ~ iPhone 17 series) without issues. However, SpeechTranscriber.isAvailable value is false for my iPhone 11 Pro. https://developer.apple.com/documentation/speech/speechtranscriber/isavailable I'am curious why the iPhone 11 Pro device is not supported. Are all iPhone 11 series not supported intentionally? Or is there any problem with my specific device? I've also checked the supportedLocales, and the value is an empty array. https://developer.apple.com/documentation/speech/speechtranscriber/supportedlocales

I have the new iOS 26 SpeechTranscriber working in my application. The issue I am facing is how to determine if the device I am running on supports SpeechTranscriber. I was able to create code that tests if the device supports transcription but it takes a bit of time to run and thus the results are not available when the app launches. What I am looking for is a list of what iOS 26 devices it doesn't run on. I think its safe to assume any new devices will support it so if we can just have a list of what devices that can run iOS 26 and not able to do transcription it would be much faster for the app. I have determined it doesn't work on a SE 2nd Gen, it works on iPhone 12, SE 3rd Gen, iPhone 14 Pro, 15 Pro. As the SpeechTranscriber doesn't work in the simulator I can't determine that way. I have checked the docs and it doesn't list the devices it doesn't work on.

Hello All, I want to implement Text-to-Speech (TTS) and vibration functionality when a push notification arrives. In my app, I am already using Critical Alerts, and the critical alert sound plays correctly in all app states. However, I need to confirm whether it is possible to trigger Text-to-Speech and custom vibration in all app states: Foreground Background Terminated (killed) state My Questions: Is it technically possible for iOS to run Text-to-Speech (using AVSpeechSynthesizer) when a critical alert notification arrives in background or terminated state? Is it possible to trigger custom vibration patterns from a critical alert when the app is not running? If yes, can someone please provide guidance or sample code on how to implement this? If no, can Apple explain the limitation or provide documentation confirming that TTS and vibration cannot be triggered in background/kill states? What works currently: TTS and vibration only work in foreground when the app is active. Critical alert sound works correctly in all states. I want confirmation on whether iOS supports background/terminated TTS and vibration, or if this is a platform restriction even when using Critical Alerts. Thank you!

I'm using the new SpeechAnalyzer framework to detect certain commands and want to improve accuracy by giving context. Seems like AnalysisContext is the solution for this, but couldn't find any usage example. So I want to make sure that I'm doing it right or not. let context = AnalysisContext() context.contextualStrings = [ AnalysisContext.ContextualStringsTag("commands"): [ "set speed level", "set jump level", "increase speed", "decrease speed", ... ], AnalysisContext.ContextualStringsTag("vocabulary"): [ "speed", "jump", ... ] ] try await analyzer.setContext(context) With this implementation, it still gives outputs like "Set some speed level", "It's speed level", etc. Also, is it possible to make it expect number after those commands, in order to eliminate results like "set some speed level to" (instead of two).

Hello, I am building an iOS-only, commercial app that uses AVSpeechSynthesizer with system voices, strictly using the APIs provided by Apple. Before distributing the app, I want to ensure that my current implementation does not conflict with the iOS Software License Agreement (SLA) and is aligned with Apple’s intended usage. For a better playback experience (more accurate estimation of utterance duration and smoother skip forward/backward during playback), I currently synthesize speech using: AVSpeechSynthesizer.write(_:toBufferCallback:) Converting the received AVAudioPCMBuffer buffers into audio data Storing the audio inside the app sandbox Playing it back using AVAudioPlayer / AVAudioEngine The cached audio is: Generated fully on-device using system voices Stored only inside the app’s private container Used only for internal playback controls (timeline, seek, skip ±5 seconds) Never shared, exported, uploaded, or exposed outside the app The alternative approaches would be: Keeping the generated audio entirely in memory (RAM) for playback purposes, without writing it to the file system at any point Or using AVSpeechSynthesizer.speak(_:) and playing speech strictly in real time which has a poorer user experience compared to my approach I have reviewed the current iOS Software License Agreement: https://www.apple.com/legal/sla/docs/iOS18_iPadOS18.pdf In particular, section (f) mentions restrictions around System Characters, Live Captions, and Personal Voice, including the following excerpt: “…use … only for your personal, non-commercial use… No other creation or use of the System Characters, Live Captions, or Personal Voice is permitted by this License, including but not limited to the use, reproduction, display, performance, recording, publishing or redistribution in a … commercial context.” I do not see a specific reference in the SLA to system text-to-speech voices used via AVSpeechSynthesizer, and I want to be certain that temporarily caching synthesized speech for internal, non-exported playback is acceptable in a commercial app. My question is: Is caching AVSpeechSynthesizer system-voice output inside the app sandbox for internal playback acceptable, or is Apple’s recommended approach to rely only on real-time playback (speak(_:)) or strictly in-memory buffering without file storage? If this question falls outside DTS technical scope and is instead a policy or licensing matter, I would appreciate guidance on the authoritative Apple documentation or the correct Apple team/contact. Thank you.

I am a graduate student conducting research in speech/audio signal processing and multimodal interaction. Apple Vision Pro is widely recognized as a multimodal interactive system supporting voice, eye, and gesture inputs. However, I could not find detailed specifications or documentation about the audio input sampling rate used by the device’s built-in microphone array when capturing user audio. Specifically, I would like to understand: What is the default audio input sampling rate (e.g., 16 kHz, 44.1 kHz, 48 kHz, etc.) for the Vision Pro’s microphones? When developing with visionOS / AVAudioSession / AVAudioEngine, is there a documented or recommended sampling rate for audio capture? Are there any best practices or settings for enabling high-quality voice capture on Vision Pro (especially for voice research tasks)? For context, my work involves voice processing, analysis, and possibly on-device real-time speech recognition. Any pointers to relevant APIs, documentation or examples (especially regarding audio capture buffer size or available formats on visionOS) would be very helpful. Thank you in advance! Best regards.

Hello, I want to use the Speech framework in my app. However, I found that if I want it to work offline, it must be downloaded separately on the device. Do I understand correctly that it is not allowed to use it in a Swift Student Challenge submission if English (as the speech language) must be downloaded by the tester on their device using the internet beforehand?

Is it possible to perform speech-to-text using AVAudioEngine to capture microphone input while being on a FaceTime call at the same time? I tried implementing this, but whenever I attempt to start the  AVAudioEngine  while a FaceTime call is active, I get the following error: “The operation couldn’t be completed. (OSStatus error 2003329396)” I assume this might be due to microphone resource restrictions during FaceTime, but I’d like to confirm whether this limitation is at the system level or if there’s any possible workaround or entitlement that allows concurrent microphone access. Has anyone encountered this issue or found a solution?

I'm trying to use the new Speech framework for streaming transcription on macOS 26.3, and I can reproduce a failure with SpeechAnalyzer.start(inputSequence:). What is working: SpeechAnalyzer + SpeechTranscriber offline path using start(inputAudioFile:finishAfterFile:) same Spanish WAV file transcribes successfully and returns a coherent final result What is not working: SpeechAnalyzer + SpeechTranscriber stream path using start(inputSequence:) same WAV, replayed as AnalyzerInput(buffer:bufferStartTime:) fails once replay starts with: _GenericObjCError domain=Foundation._GenericObjCError code=0 detail=nilError I also tried: DictationTranscriber instead of SpeechTranscriber no realtime pacing during replay Both still fail in stream mode with the same error. So this does not currently look like a ScreenCaptureKit issue or a Python integration issue. I reduced it to a pure Swift CLI repro. Environment: macOS 26.3 (25D122) Xcode 26.3 Swift 6.2.4 Apple Silicon Mac Has anyone here gotten SpeechAnalyzer.start(inputSequence:) working reliably on macOS 26.x? If so, I'd be interested in any workaround or any detail that differs from the obvious setup: prepareToAnalyze(in:) bestAvailableAudioFormat(...) AnalyzerInput(buffer:bufferStartTime:) replaying a known-good WAV in chunks I already filed Feedback Assistant: FB22149971

We built an open-source macOS menu bar app that turns speech into text and pastes it into the active app — using SpeechAnalyzer for on-device transcription, ScreenCaptureKit + Vision for screen-aware context, and FluidAudio for speaker diarization in meeting mode. Here's what we learned shipping it on macOS 26. GitHub: github.com/Marvinngg/ambient-voice Architecture The app has two modes: hotkey dictation (press to talk, release to inject) and meeting recording (continuous transcription with a floating panel). Dictation Mode Audio capture uses AVCaptureSession (more on why below). The captured audio feeds into SpeechAnalyzer via an AsyncStream: let transcriber = SpeechTranscriber( locale: locale, transcriptionOptions: [], reportingOptions: [.volatileResults, .alternativeTranscriptions], attributeOptions: [.audioTimeRange, .transcriptionConfidence] ) let analyzer = SpeechAnalyzer(modules: [transcriber]) let (inputSequence, inputBuilder) = AsyncStream.makeStream() try await analyzer.start(inputSequence: inputSequence) While recording, we capture a screenshot of the focused window using ScreenCaptureKit, run Vision OCR (VNRecognizeTextRequest), extract keywords, and inject them into SpeechAnalyzer as contextual bias: let context = AnalysisContext() context.contextualStrings[.general] = ocrKeywords try await analyzer.setContext(context) This improves accuracy for technical terms and proper nouns visible on screen. If your screen shows "SpeechAnalyzer", saying it out loud is more likely to be transcribed correctly. After transcription, an optional L2 step sends the text through a local LLM (ollama) for spoken-to-written cleanup, then CGEvent simulates Cmd+V to paste into the active app. Meeting Mode Meeting mode forks the same audio stream to two consumers: SpeechAnalyzer — real-time streaming transcription, displayed in a floating NSPanel FluidAudio buffer — accumulates 16kHz Float32 mono samples for batch speaker diarization after recording stops When the user ends the meeting, FluidAudio's performCompleteDiarization() runs on the accumulated audio. We align transcription segments with speaker segments using audioTimeRange overlap matching — each transcription segment gets assigned the speaker ID with the most time overlap. Results export to Markdown. Pitfalls We Hit on macOS 26 1. AVAudioEngine installTap doesn't fire with Bluetooth devices We started with AVAudioEngine.inputNode.installTap() for audio capture. It worked fine with built-in mics but the tap callback never fired with Bluetooth devices (tested with vivo TWS 4 Hi-Fi). Fix: switched to AVCaptureSession. The delegate callback captureOutput(_:didOutput:from:) fires reliably regardless of audio device. The tradeoff is you get CMSampleBuffer instead of AVAudioPCMBuffer, so you need a conversion step. 2. NSEvent addGlobalMonitorForEvents crashes Our global hotkey listener used NSEvent.addGlobalMonitorForEvents. On macOS 26, this crashes with a Bus error inside GlobalObserverHandler — appears to be a Swift actor runtime issue. Fix: switched to CGEventTap. Works reliably, but the callback runs on a CFRunLoop context, which Swift doesn't recognize as MainActor. 3. CGEventTap callbacks aren't on MainActor If your CGEventTap callback touches any @MainActor state, you'll get concurrency violations. The callback runs on whatever thread owns the CFRunLoop. Fix: bridge with DispatchQueue.main.async {} inside the tap callback before touching any MainActor state. 4. CGPreflightScreenCaptureAccess doesn't request permission We used CGPreflightScreenCaptureAccess() as a guard before calling ScreenCaptureKit. If it returned false, we'd bail out. The problem: this function only checks — it never triggers macOS to add your app to the Screen Recording permission list. Chicken-and-egg: you can't get permission because you never ask for it. Fix: call CGRequestScreenCaptureAccess() at app startup. This adds your app to System Settings → Screen Recording. Then let ScreenCaptureKit calls proceed without the preflight guard — SCShareableContent will also trigger the permission prompt on first use. 5. Ad-hoc signing breaks TCC permissions on every rebuild During development, codesign --sign - (ad-hoc) generates a different code directory hash on every build. macOS TCC tracks permissions by this hash, so every rebuild = new app identity = all permissions reset. Fix: sign with a stable certificate. If you have an Apple Development certificate, use that. The TeamIdentifier stays constant across rebuilds, so TCC permissions persist. We also discovered that launching via open WE.app (LaunchServices) instead of directly executing the binary is required — otherwise macOS attributes TCC permissions to Terminal, not your app. Benchmarks We ran end-to-end benchmarks on public datasets (Mac Mini M4 16GB, macOS 26): Transcription (SpeechAnalyzer, AliMeeting Chinese): • Near-field CER 34% (excluding outliers ~25%) • Far-field CER 40% (single channel, no beamforming, >30% overlap) • Processing speed 74-89x real-time Speaker diarization (FluidAudio offline): • AMI English 16 meetings: avg DER 23.2% (collar=0.25s, ignoreOverlap=True) • AliMeeting Chinese 8 meetings: DER 48.5% (including overlap regions) • Memory: RSS ~500MB, peak 730-930MB Full evaluation methodology, scripts, and raw results are in the repo. Open Source The project is MIT licensed: github.com/Marvinngg/ambient-voice It includes the macOS client (Swift 6.2, SPM), server-side distillation/training scripts (Python), and a complete evaluation framework with reproducible benchmarks. Feedback and contributions welcome.

We built an open-source macOS menu bar app that turns speech into text and pastes it into the active app — using SpeechAnalyzer for on-device transcription, ScreenCaptureKit + Vision for screen-aware context, and FluidAudio for speaker diarization in meeting mode. Here's what we learned shipping it on macOS 26. GitHub: github.com/Marvinngg/ambient-voice Architecture The app has two modes: hotkey dictation (press to talk, release to inject) and meeting recording (continuous transcription with a floating panel). Dictation Mode Audio capture uses AVCaptureSession (more on why below). The captured audio feeds into SpeechAnalyzer via an AsyncStream: let transcriber = SpeechTranscriber( locale: locale, transcriptionOptions: [], reportingOptions: [.volatileResults, .alternativeTranscriptions], attributeOptions: [.audioTimeRange, .transcriptionConfidence] ) let analyzer = SpeechAnalyzer(modules: [transcriber]) let (inputSequence, inputBuilder) = AsyncStream.makeStream() try await analyzer.start(inputSequence: inputSequence) While recording, we capture a screenshot of the focused window using ScreenCaptureKit, run Vision OCR (VNRecognizeTextRequest), extract keywords, and inject them into SpeechAnalyzer as contextual bias: let context = AnalysisContext() context.contextualStrings[.general] = ocrKeywords try await analyzer.setContext(context) This improves accuracy for technical terms and proper nouns visible on screen. If your screen shows "SpeechAnalyzer", saying it out loud is more likely to be transcribed correctly. After transcription, an optional L2 step sends the text through a local LLM (ollama) for spoken-to-written cleanup, then CGEvent simulates Cmd+V to paste into the active app. Meeting Mode Meeting mode forks the same audio stream to two consumers: SpeechAnalyzer — real-time streaming transcription, displayed in a floating NSPanel FluidAudio buffer — accumulates 16kHz Float32 mono samples for batch speaker diarization after recording stops When the user ends the meeting, FluidAudio's performCompleteDiarization() runs on the accumulated audio. We align transcription segments with speaker segments using audioTimeRange overlap matching — each transcription segment gets assigned the speaker ID with the most time overlap. Results export to Markdown. Pitfalls We Hit on macOS 26 1. AVAudioEngine installTap doesn't fire with Bluetooth devices We started with AVAudioEngine.inputNode.installTap() for audio capture. It worked fine with built-in mics but the tap callback never fired with Bluetooth devices (tested with vivo TWS 4 Hi-Fi). Fix: switched to AVCaptureSession. The delegate callback captureOutput(_:didOutput:from:) fires reliably regardless of audio device. The tradeoff is you get CMSampleBuffer instead of AVAudioPCMBuffer, so you need a conversion step. 2. NSEvent addGlobalMonitorForEvents crashes Our global hotkey listener used NSEvent.addGlobalMonitorForEvents. On macOS 26, this crashes with a Bus error inside GlobalObserverHandler — appears to be a Swift actor runtime issue. Fix: switched to CGEventTap. Works reliably, but the callback runs on a CFRunLoop context, which Swift doesn't recognize as MainActor. 3. CGEventTap callbacks aren't on MainActor If your CGEventTap callback touches any @MainActor state, you'll get concurrency violations. The callback runs on whatever thread owns the CFRunLoop. Fix: bridge with DispatchQueue.main.async {} inside the tap callback before touching any MainActor state. 4. CGPreflightScreenCaptureAccess doesn't request permission We used CGPreflightScreenCaptureAccess() as a guard before calling ScreenCaptureKit. If it returned false, we'd bail out. The problem: this function only checks — it never triggers macOS to add your app to the Screen Recording permission list. Chicken-and-egg: you can't get permission because you never ask for it. Fix: call CGRequestScreenCaptureAccess() at app startup. This adds your app to System Settings → Screen Recording. Then let ScreenCaptureKit calls proceed without the preflight guard — SCShareableContent will also trigger the permission prompt on first use. 5. Ad-hoc signing breaks TCC permissions on every rebuild During development, codesign --sign - (ad-hoc) generates a different code directory hash on every build. macOS TCC tracks permissions by this hash, so every rebuild = new app identity = all permissions reset. Fix: sign with a stable certificate. If you have an Apple Development certificate, use that. The TeamIdentifier stays constant across rebuilds, so TCC permissions persist. We also discovered that launching via open WE.app (LaunchServices) instead of directly executing the binary is required — otherwise macOS attributes TCC permissions to Terminal, not your app. Benchmarks We ran end-to-end benchmarks on public datasets (Mac Mini M4 16GB, macOS 26): Transcription (SpeechAnalyzer, AliMeeting Chinese): • Near-field CER 34% (excluding outliers ~25%) • Far-field CER 40% (single channel, no beamforming, >30% overlap) • Processing speed 74-89x real-time Speaker diarization (FluidAudio offline): • AMI English 16 meetings: avg DER 23.2% (collar=0.25s, ignoreOverlap=True) • AliMeeting Chinese 8 meetings: DER 48.5% (including overlap regions) • Memory: RSS ~500MB, peak 730-930MB Full evaluation methodology, scripts, and raw results are in the repo. Open Source The project is MIT licensed: github.com/Marvinngg/ambient-voice It includes the macOS client (Swift 6.2, SPM), server-side distillation/training scripts (Python), and a complete evaluation framework with reproducible benchmarks. Feedback and contributions welcome.