How to Add Voice and Speech-to-Text to an iPhone App (2026)
Add voice to an iPhone app without code: record memos, turn speech into text and read text aloud with Apple's own frameworks, step by step with Borel.
By Sigsten Gustavii· · 8 min read
To add speech-to-text to an iPhone app, decide whether people dictate live or transcribe recordings, ask for the microphone and speech recognition permissions with clear reasons when the feature is used, use Apple's Speech framework to turn audio into text, and let people edit the result. With Borel, you describe the voice feature and it writes the Swift.
| Feature | Apple tool | Permission | Costs to run |
|---|---|---|---|
| Record a voice memo | AVFoundation (AVAudioRecorder) | Microphone | Nothing |
| Speech to text | Speech framework (SFSpeechRecognizer) | Microphone and speech recognition | Nothing |
| Read text aloud | AVSpeechSynthesizer | None | Nothing |
| Play back audio | AVFoundation | None | Nothing |
| The keyboard's own dictation | Built into iOS text fields | Handled by iOS | Nothing |
Voice makes an app usable when hands are busy or typing is slow: in the kitchen, on a walk, in a meeting, for people who find small keyboards hard. And on iPhone it is all built in. There is no transcription service to sign up for and no per-minute bill.
Three voice features, three frameworks
Recording. AVFoundation records into the app's own files, after iOS asks for the microphone. The app can list memos with their length, play them back and pause. Speech to text. Apple's Speech framework turns spoken words into text, live while someone talks or from a recording, with its own permission. Text to speech. AVSpeechSynthesizer reads any text in the app aloud with the iPhone's built-in voices, with no permission at all.
App ideas with voice
| App | Voice feature | Why |
|---|---|---|
| Voice memo app | Record, list, play, transcribe | Capture ideas in seconds |
| Spoken journal | Speech to text | Talk about your day, search it later |
| Meeting notes | Record and transcribe | Read a meeting back as text |
| Recipe app | Read steps aloud | Hands covered in flour |
| Bedtime stories | Read aloud, page by page | A calm voice at night |
| Language practice | Say a phrase, see what the phone heard | Instant pronunciation check |
| Field reports | Dictate notes with photos | Faster than typing outdoors |
More ideas on Borel's voice and speech to text page.
Designing for the ear is different from designing for the eye, and the biggest change is pace. Someone dictating doesn't look at the screen much, so confirm with a sound or a light haptic when recording starts and stops, and keep the important buttons large enough to hit without looking. For read-aloud, show which sentence is being spoken, so people can follow along or find their place after a pause.
Permissions, explained well
Speech to text needs two permissions: the microphone and speech recognition. Recording alone needs only the microphone, and reading aloud needs none. Apple's Guideline 5.1.1(ii) asks for purpose sentences that clearly say how the app uses them. Borel writes both for each app and fills them in if they would ever be missing. Ask at the moment someone taps the microphone, not when the app opens, and keep the app useful when they say no: typing should always work.
| Permission | A clear sentence |
|---|---|
| Microphone | "Record voice memos you can play back and turn into text." |
| Speech recognition | "Turn what you say into text for your journal." |
On the device or in the cloud?
Apple's speech recognition can run on the iPhone itself for supported languages, so audio doesn't have to leave the phone, and it works offline. Where on-device recognition isn't available, Apple's servers may handle it. Either way, it is Apple's service, not a third-party AI, and it costs your app nothing. If you later send a transcript to Borel AI to summarise it, that is a third-party AI request, and Apple's Guideline 5.1.2(i) consent applies; Borel's AI kit asks for it.
Design tips for voice
- One big microphone button, with a clear recording state and a timer.
- Show the words as they come during live dictation, so people know it is working.
- Always let people edit the text afterwards; recognition is good, not perfect.
- Keep the recording alongside the transcript when accuracy matters.
- Offer read-aloud speeds, and pause and resume.
- Choose the language if your app is used in more than one.
Live dictation or recorded memos?
| Live dictation | Record, then transcribe | |
|---|---|---|
| How it feels | Words appear as you speak | Speak freely, read the text afterwards |
| Good for | Short notes, search, messages | Long memos, meetings, journals |
| Keeps the audio | Not unless you ask | Yes, alongside the text |
| If recognition slips | Fix it on the spot | Listen again to check |
| Length | Short bursts | Longer recordings, transcribed after |
Many apps offer both: a quick dictation button in text fields and a separate recorder for longer thoughts. Remember that the keyboard's own dictation button already works in any text field of a native app, so a dedicated voice feature should do something more, such as keeping the recording, transcribing it later or reading text back.
Accessibility and voice
Voice features help many people beyond those in a hurry. Reading text aloud helps people with low vision or dyslexia, and dictation helps people who find typing hard. Native SwiftUI apps also work with VoiceOver, Apple's screen reader, out of the box. Give every voice button a clear label, such as "Start recording" rather than an unlabelled microphone, so VoiceOver can announce it.
Testing voice
The iOS Simulator in your browser runs the screens, but speaking and listening need a real iPhone. Install the app with Borel Companion, which comes with Pro, and test in a quiet room and a noisy one, with short and long recordings, and with permissions denied. Check that read-aloud stops when another app starts playing.
Apple's rules for voice features
| Guideline | What Apple asks | In a Borel app |
|---|---|---|
| 5.1.1(ii) Permission | Clear purpose sentences for microphone and speech | Written for your app and filled in if missing |
| 5.1.1(iii) Data minimisation | Only what the feature needs | Recording only when the person taps |
| 5.1.2(i) Third-party AI | Consent before data goes to an AI | Asked if a transcript is sent to Borel AI |
| 5.1.1(i) Privacy policy | Say what you collect | Borel hosts a policy for the app |
Read the full text on Apple's App Review Guidelines page.
How to choose the right way to add voice
- Borel, for recording, speech to text and read-aloud in a native iPhone app from a sentence, with permissions handled.
- The Speech framework by hand in Xcode, if you are a Swift developer and need custom vocabularies or fine control.
- A cloud transcription service, if you need speaker labels for long meetings; add its key under Tools, in Keys, and test it carefully.
How to do it with Borel, step by step
Step 1: Describe the voice feature
Add voice to my journal. A big microphone button records what I say and shows the words live as I talk; when I stop, the text goes into a new entry I can edit, and the recording is kept with it. Each entry has a Play button for the recording and a Read aloud button for the text. Ask for the microphone and speech permissions only when I tap the button, and keep typing working if I say no.You can also choose Microphone from Device in the chat's + menu. Rivet writes it with AVFoundation, the Speech framework and AVSpeechSynthesizer.
Step 2: Check the screens in the simulator
On the live iPhone in your browser, check the recording states, the entry list and what happens when permission is refused.
Step 3: Talk to it on your iPhone
Install the app with Companion (Pro) and use it for real: a long entry, a short one, a noisy café.
Step 4: Publish
Stamp checks that every permission has a clear reason, and Publish builds and uploads the app under your Apple Developer account.
The Bottom Line
On iPhone, voice is built in: AVFoundation records, the Speech framework transcribes, and the system voices read aloud, with no AI account and no running cost. Ask for permission when someone taps the microphone, let them edit what was heard, and test on a real phone. Borel writes it all from a sentence. Open your app in Borel and say something to it.
Questions people ask
How do I add speech-to-text to an iPhone app?
Use Apple's Speech framework with microphone and speech recognition permissions. In Borel, describe the feature or pick Microphone from Device in the + menu, and it writes the Swift.
Does speech-to-text cost money?
Not with Apple's Speech framework: there is no per-minute charge to the app. It isn't Borel's AI, so it doesn't use your app's balance.
Does speech recognition work offline?
For supported languages, Apple's recognition can run on the device and work offline. Otherwise it may use Apple's servers.
Can an app read text aloud?
Yes, with AVSpeechSynthesizer and the iPhone's built-in voices, with no permission or key.
Can I test voice in the simulator?
You can test the screens; speaking and listening need a real iPhone. Install the app with Borel Companion, which comes with Pro.
Can the app transcribe recordings later?
Yes. The Speech framework can transcribe a saved recording as well as live speech.
Can AI summarise a transcript?
Yes. Send the text to Borel AI with an instruction like "three bullet points". The app asks for consent before the opening AI request, as Apple requires.
Where are recordings stored?
In the app's own files on the iPhone, or, with sign-in, uploaded to the cloud Borel runs for your app so they show on other devices.
Can people use voice without giving microphone access?
No: recording and speech to text need the microphone. Read-aloud needs no permission, and typing should always work as a fallback.
Can I add voice to an app I already built with Borel?
Yes. Ask for it in the app's chat, for example "add a microphone button that turns speech into a new note", and Borel adds it as a new version you can roll back.
Is speech recognition accurate enough for meetings?
For clear speech in a quiet room it does well; with crosstalk and noise it slips. Keep the recording with the transcript so people can check what was said.
Which languages are supported?
Apple's speech recognition supports many languages, with on-device support for a subset. Let people choose the language if your app is used in more than one.
Voice is built into every iPhone, and I think more apps should listen. Say something to yours and see what it hears.
– Sigsten
Make your iPhone app in Borel
Describe the app you want and see its first screens before you pay anything.
See your app free