The problem
Most language apps teach reading and tapping. Speaking practice is what learners need and what apps do worst, because a voice conversation with an AI falls apart if there is lag, echo or no way to interrupt.
What we built
A native iOS app that streams microphone audio to our server over a WebSocket and plays the tutor's reply as it arrives. Barge-in stops the tutor the moment the learner speaks. Each session is guided by a lesson goal from a structured course, and scores for pronunciation, fluency and grammar are stored so progress is visible.
Why it matters for clients
Realtime voice is one of the most demanding things you can put in an app. The same approach works for voice ordering, support agents, interview practice and accessibility tools.
What is inside
- Native iOS app in Swift for low-latency audio
- Full-duplex voice: the learner can interrupt the tutor mid-sentence
- Hardware echo cancellation so the tutor does not hear itself
- Our own server relays the realtime voice session, so the AI provider key never sits on the phone
- A structured course of units and lessons, not an open-ended chatbot
Technology
Swift (iOS), Realtime voice AI, WebSockets, Node.js, PostgreSQL.