Real-time German Language Tutor

German voice tutor that interacts with the user in real-time, gives pronunciation and grammar feedback, and keeps track of lesson progress. How it works (high-level, API-based): 1. User input: * Accept speech or text via a WebRTC stream. * Stream audio in small chunks for fast response. 2. Speech-to-text (STT): * Convert user speech to text using an API like Deepgram or AssemblyAI. * Get word-level timestamps and confidence scores. 3. Session manager / database: * Track lesson, scenario, user progress, errors, and previous tips. * Flag mispronounced or low-confidence words. * Store test results and use them to guide future lessons. 4. Pronunciation check: * Use a small LLM (like GPT-4o-mini) to generate simple pronunciation tips for flagged words. 5. Conversation & correction: * Another LLM (GPT-5 or GPT-5-mini) takes the user input, session info, and pronunciation tips to: * Suggest corrected phrasing * Give grammar or pronunciation hints * Generate the next conversational prompt 6. Text-to-speech (TTS): * Convert the AI’s response to natural-sounding German audio using Cartesia Sonic API. * Stream it back to the user via WebRTC. Summary: Essentially, it’s just connecting a few APIs (STT, LLMs, TTS) with a database to track lessons and user progress. No heavy AI model training is needed—the intelligence comes from the APIs. The coder mainly needs to handle: audio streaming, API integration, session state, and result display.

Project ID: 40182450