English Pronunciation Scoring Model

Build a production-ready machine-learning system that evaluates a learner’s spoken English by comparing microphone audio against expected text and returns real-time pronunciation feedback, including numeric accuracy scores at word and sentence level and a simple quality label (good / poor), with support for both isolated words and connected speech; the solution must be low-latency, scalable, and deployable as an inference service suitable for mobile app integration—please review the attached document for the complete functional requirements, scoring logic, pronunciation flow, and end-to-end project architecture.