distortions in the air traveling to the microphone of the language model trained to pick it up, could cause “odd results” lol