I need a working firmware prototype that turns the ESP32-Korvo V2 into a fully offline voice interface for a motorized wheelchair. Using the on-board microphone array and ESP-DSP / ESP-SR libraries, the system must: • Recognise four English commands—“forward,” “backward,” “left,” and “right”—locally, with no cloud calls and minimal latency. • Light a dedicated LED for each command; I’ll define the exact colour mapping, so please make the pin assignments configurable. • Detect unknown or unclear speech and activate a separate “error” LED so I can spot misfires immediately. Please deliver: 1. Source code (Arduino IDE or ESP-IDF) with clear comments. 2. Model training scripts or steps so I can retrain or add commands later. 3. Pinout diagram and a brief setup guide that lets me flash, test, and adjust LED colours quickly. Keep RAM/flash usage reasonable, and feel free to recommend lightweight wake-word or noise-rejection techniques to improve accuracy in a typical indoor environment.