Skip to content

Dictation for iPhone · Fully on-device

Speak Turkish.
Keep the English words in English.

TrenVoice is built for people who mix Turkish and English in one sentence. Say "production'daki endpoint'i deploy ettim" and that is exactly what appears on screen — not a phonetic guess at it.

Coming soon to the App StoreSee the difference →
  • No subscription
  • No word limit
  • No account
TrenVoiceListening

Bugün production'daki endpoint'i deploy ettim ama Cloudflare rate limit yüzünden request'ler fail oluyor.

Processed on deviceInsert

The problem shows up in a single sentence

In code-switched speech, Turkish dictation spells English words the way it hears them. The result is manual cleanup after every message. TrenVoice treats these sentences as the normal case, not the exception.

What you said

Bugün production'daki endpoint'i deploy ettim ama Cloudflare rate limit yüzünden request'ler fail oluyor.

What you are used to

Bugün prodakşındaki end pointi diploy ettim ama klaud fleyr reyt limit yüzünden rikuestler feyl oluyor.

TrenVoice

Bugün production'daki endpoint'i deploy ettim ama Cloudflare rate limit yüzünden request'ler fail oluyor.

Illustrative examples. Measured comparative accuracy results will be published closer to launch, once they have been run against our own held-out test set.

Three design decisions

Two languages, one sentence

Turkish suffixes stay correct on English stems: endpoint'i, GitHub'daki, API'ye. Meaning is never changed and your sentence is never rewritten — it is just spelled correctly.

Audio never leaves the phone

The speech model is downloaded once; everything after that runs on device. It works in airplane mode. No audio, text or usage data is sent anywhere.

An app, not a subscription

No word quota, no monthly minutes, no per-dictation charge. Transcription runs on your own hardware, so there is no recurring cost behind it.

Record, review, insert

Three steps, no surprises — nothing is typed anywhere until you say so.

  1. 01

    Record

    Start from the app, the Action Button or a Shortcut. You get an unambiguous "listening" indicator; while the model is still loading it says "preparing" instead.

  2. 02

    Review

    The raw transcript and the glossary-corrected version sit side by side. Any correction you dislike can be reverted individually.

  3. 03

    Insert

    Switch to the TrenVoice keyboard in WhatsApp, Teams or Mail and tap Insert. The text lands at your cursor — nothing is auto-sent and no Return is simulated.

Why does recording start in the app?

iOS does not allow custom keyboards to access the microphone. That is Apple's platform rule, not our design choice. So recording happens in the app and the result is inserted from the keyboard with one tap. Nothing is ever pasted automatically. In places where iOS mandates the system keyboard — password fields, for example — this flow does not apply and the system keyboard takes over.

In apps that accept custom keyboards

  • WhatsApp
  • Microsoft Teams
  • Slack
  • Mail
  • Notes
  • Telegram

Your own technical glossary

Add the terms you actually use and TrenVoice spells them the same way every time, with the right capitalisation and the right Turkish suffix. The glossary is entirely yours: you add it, you delete it, and it never leaves your device.

Corrections arrive as suggestions and can be reverted. A rule you have not approved is never applied. Numbers, negations, URLs, version strings and code-like tokens are never "corrected" — no change that would alter the meaning of a sentence is made silently.

Example seeded terms

  • GitHub
  • Cloudflare
  • Supabase
  • PostgreSQL
  • TypeScript
  • OpenAI
  • pnpm
  • API
  • webhook
  • endpoint
  • deploy
  • production
  • pull request
  • commit
  • meeting
  • deadline
  • feedback
  • check-in

Suffixes survive

  • GitHub'daki
  • API'ye
  • endpoint'i
  • branch'ini
  • production'da
  • deadline'ı

Privacy is the architecture, not a setting

These are consequences of how the app is built, not marketing promises.

  • Audio never leaves the device

    Your speech is held in memory only for the duration of the session and released when it ends. Keeping a recording is a separate, explicit action.

  • No account, no sign-in

    No email, no phone number, no username. There is nothing to sign up for.

  • Offline after setup

    Once the model is installed, the dictate–correct–insert path makes no network requests at all. It is tested in airplane mode.

  • History is off by default

    If you do turn it on, transcripts stay on your device and can be deleted completely whenever you want.

  • The keyboard does not collect what you type

    It records no keystroke history and makes no attempt to identify which app you are in.

  • No analytics SDK

    No third-party trackers, no advertising identifiers, no behavioural telemetry.

The only network traffic is the initial speech-model download. During that download ordinary connection metadata such as your IP address is visible to the model host; no audio or text is sent. You can also install the model from your own local file instead.

Where we sit

The practical differences between on-device dictation and cloud dictation services.

Where we sit
TrenVoiceBuilt-in iOS dictationCloud dictation apps
Turkish + English in one sentencePrimary design goalLimitedVaries
Works offlineYes (after setup)Device-dependentNo
Audio sent to a serverNoDevice-dependentYes
Account requiredNoNoUsually yes
Monthly subscriptionNoNoUsually yes
Word or minute quotaNoNoCommon
Personal technical glossaryYesNoSometimes

Based on publicly available product information and our own daily use. Apple's on-device dictation behaviour varies by device, language and iOS version, and cloud services differ by provider.

Frequently asked questions

What is TrenVoice?

A dictation app for iPhone with a companion keyboard, designed to correctly spell the English technical and everyday words that appear inside Turkish speech. It converts speech to text; it does not read text aloud, translate, or record meetings.

How does it get mixed Turkish-English sentences right?

Two layers work together: a multilingual speech model running on device, and your own approved glossary of terms. The glossary only touches terms you added and the Turkish suffixes attached to them; the rest of the sentence is left exactly as spoken.

Is my audio or text sent to a server?

No. Dictation runs entirely on your device. The only network traffic is the initial speech-model download; audio and text are never uploaded.

Does it work without internet?

Yes. Once the model is installed, recording, transcription, correction and insertion all work in airplane mode.

Can I dictate from inside the keyboard itself?

iOS does not permit custom keyboards to use the microphone. Recording therefore starts in the app — from the app itself, the Action Button or a Shortcut. The result appears in the keyboard and lands at your cursor when you tap Insert.

Which apps does it work in?

Apps that accept custom keyboards: WhatsApp, Teams, Slack, Mail, Notes and similar. It does not work in fields where iOS mandates the system keyboard, such as password fields, or in apps that reject custom keyboards outright. We will publish the list of apps we have tested at launch.

Is there a subscription or a word limit?

No. Because recognition runs on your device there is no word quota, no monthly cap and no per-use charge. Pricing will be announced closer to launch.

Which iPhones are supported?

iPhones running iOS 26 or later. The on-device model needs meaningful compute and memory, so it performs noticeably better on recent iPhones. The exact supported list will be published at launch.

When is it launching?

No date yet. The app is currently being tested on physical devices while the speech model's Turkish-English performance is measured against our own test set.

Coming soon to the App Store

TrenVoice is being tested on physical devices right now. We will announce the date on this page.

iPhone · iOS 26+