BabyTalk enables offline speech-to-text and text-to-speech on ESP32 microcontrollers without cloud dependency, supporting English transcription and neural voice synthesis. The project runs on ESP32-S3 or ESP32-P4 with minimal memory requirements and works across C, MicroPython, and Erlang/Elixir, making it suitable for offline voice applications like LoRa radios and IoT devices.