99 Languages, Zero Internet: How On-Device Transcription Actually Works
Whisper Large V3 Turbo can transcribe 99+ languages entirely on your iPhone — no internet, no cloud, no data leaving your device. Here's the technology that makes it possible.
Your iPhone has a Neural Engine capable of 17 trillion operations per second. That's enough processing power to run sophisticated speech recognition models that understand 99+ languages — entirely offline, entirely on-device.
This isn't science fiction. It's what happens when models like Whisper Large V3 Turbo are optimized for mobile hardware using frameworks like CoreML and ONNX Runtime.
The Technology Stack
Whisper Large V3 Turbo is OpenAI's multilingual speech recognition model, designed to understand speech in virtually any language. The "Turbo" variant is optimized for speed without sacrificing accuracy, making it practical for real-time use on mobile devices.
Parakeet V3 is NVIDIA's speech recognition model optimized for 25 European languages. It's faster than Whisper for supported languages, making it ideal for real-time transcription where low latency matters.
Both models run on the iPhone's Neural Engine — a dedicated chip designed specifically for machine learning workloads. This means transcription doesn't drain your battery the way CPU-based processing would.
Why 99+ Languages Matters
The world is multilingual. Even in English-speaking countries:
- 47 million people in the US speak Spanish at home
- University lectures are increasingly delivered by professors with diverse accents
- International business means meetings in multiple languages
- Immigrants and travelers need transcription in their native language
Cloud transcription services support many languages too, but they require internet and send your audio to remote servers. On-device means you can transcribe a conversation in Mandarin on a subway in Beijing, a lecture in German at a university in Munich, or a meeting in Portuguese in São Paulo — all without connectivity.
Translation Built In
Whisper includes a translate mode that can take speech in any of its 99+ supported languages and produce an English transcript. This isn't a two-step process (transcribe then translate) — it's a single model that understands the source language and outputs English directly.
For multilingual users, this means you can:
- Attend a lecture in French and get English notes
- Record a conversation in Japanese and share the English summary
- Import an audio recording in Arabic and get a searchable English transcript
Accuracy in Practice
On-device transcription has reached a point where it's competitive with cloud services for most use cases:
- English: 95%+ accuracy for clear speech
- Major European languages: 90-95% accuracy
- Less common languages: 80-90% accuracy, improving with each model update
Custom dictionaries help with domain-specific vocabulary. If you're a medical professional, you can add terms like "epinephrine" or "myocardial infarction" so the model gets them right consistently.
The Keyboard Angle
DictaWiz includes a custom iOS keyboard with built-in dictation in any of the supported languages. This means you can dictate in Spanish in WhatsApp, switch to English for email, and use French in Notes — all without changing any system settings. The keyboard supports 16 layouts including QWERTY, QWERTZ, AZERTY, and Cyrillic.
About DictaWiz
DictaWiz — subtitles for your life. Real-time transcription, meeting recording, and AI voice productivity for iOS.
- 99+ languages on-device via Whisper Large V3 Turbo
- 25 European languages via Parakeet V3 (faster)
- Translate any language to English with Whisper translate mode
- Custom keyboard with 16 layouts and built-in dictation
- One-time purchase — no subscriptions, 100% on-device
- Download on the App Store