It cleverly relies on the Web Speech API to deliver the speaking parts, which helps keep the code size tiny. The only thing it’s missing is an occasional shadow of static drifting across the audio.