I built a local live translator because everything was cloud-only — then it turned into a whole ecosystem
Started simple: I wanted a GDPR-friendly speech translator for live events. Everything I found was cloud-only, subscription-based, and sent audio to someone else's servers. Not great when you're translating confidential meetings.
So I built LLT — runs 100% on Apple Silicon using Whisper for speech-to-text and Mistral 7B for translation. No internet needed, no data leaves the machine. Works on an M1 with 8GB RAM.
Then things escalated.
I added STT engines: Whisper (local), Google REST, Google gRPC streaming, Deepgram, AssemblyAI, Azure, and Vosk. Each pairs with a translation provider. Pick what fits your use case.
Using an iPhone as a wireless mic on a Mac requires a cable. In 2026. That bothered me enough to build LLT Mobile — a free iOS companion app that connects over Wi-Fi. But then I realized it could also work standalone with cloud services, so there's a Pro version ($12.99 US) that talks to Google, Azure etc. directly from your phone.
Then someone in a vMix forum asked how to get live subtitles into their broadcast. I read the thread and built direct vMix Title output — LLT sends translated text straight to vMix via API, shows up as a native overlay. I also made a free .xaml title template for it. And my other app Directors Plan can automate when the subtitle overlay appears and disappears in the production timeline.
I shipped the iOS app, the vMix integration, VAD for natural sentence detection, and seven STT engines.
What's in it now:
- 3 audio inputs: mic, system audio, single-app capture
- 101 target languages, auto source detection
- Smart VAD — waits for natural pauses, dramatically better translations
- 5 Whisper models (tiny to large), download on demand
- Live floating overlay, TXT transcript, SRT subtitle export
- Text-to-Speech output
- vMix Title output with free template
- Menu bar app, 5 UI languages
- 14-day free trial, $29 one-time
Tech: Swift/SwiftUI, Python backend for Whisper/Vosk, MLX for Mistral on Apple Neural Engine. macOS 13+, Apple Silicon only.
The ecosystem angle: LLT does the translation → sends to vMix as Title → Directors Plan controls when it appears in the timeline → one-click export to Stream Deck. Everything connects.
(LLT Mobile is ready to ship but currently waiting on Apple to fix a typo in my decades - old developer account — they named my team "Mathias Mathias" instead of "Mathias Schneider". Never mattered until I needed to sell something. Always fun.)
https://adelvo.io/local-live-translator/
Solo dev, self-funded, built because I needed it. Ask me anything about local AI on Apple Silicon, the STT engine landscape, or building an iOS companion app for a Mac product.
