Automatic speech recognition (ASR) running on the ESP32-S3 based on a 13.1M parameters convolution transformer.