Licences
Credits owed for what you hear in a lesson.
How your voice is handled
While a lesson is on your screen, the microphone stays on, so your phone or browser shows it as in use; it switches off when the lesson ends. Only what you say while you hold the button is recorded and sent; between presses nothing is. When you hold the button and speak, your voice is sent over the internet to a speech-recognition service so Layla can understand you. For most accounts that is Deepgram, in the United States. What you said, written out as text, and the lesson so far are then sent to Google's Gemini service, which writes Layla's reply. Layla's voice is made from that reply text by Microsoft Azure, in Singapore (or the United Arab Emirates if Singapore is unavailable or slow). Accounts we have moved onto our Arabic path are served differently while they practise Arabic. Their voice goes to Soniox, in India, instead of Deepgram; Deepgram hears it only if Soniox cannot finish. Google may run some of Layla's replies in its European region. Layla's English is made on our own server in India, not by an outside service, and her Arabic by Cartesia, which receives the Arabic parts of Layla's reply as text, never your voice. Cartesia is hosted in the United States and has no Gulf region, and we submitted its training opt-out on 13 October 2026. If Cartesia cannot make a line, our own server makes it; if our own voice cannot start a reply, Azure makes it. When we move an account onto this path, these services replace the ones above for its Arabic practice, and moving it back restores them. Our server never saves your recording. Your words and Layla's replies are kept with your lessons, in our database (Supabase) in Mumbai, India.