Question: A speech recognizer is trained on clean microphone recordings, then evaluated on a development set made from conversations captured in a moving vehicle. Explain why the score can drop sharply and name the environmental factors that are responsible.

Sample answer: The drop in accuracy comes from a shift in the recording conditions between the two sets. The training examples were collected in a controlled, low-noise setting, while the development examples were taken inside a vehicle. In that setting, background sounds such as the car’s engine, tire/road hum, and other cabin noise interfere with the speech signal, so the model performs worse because it has not learned to handle that kind of audio.

Key points:

• Training data came from a controlled, low-noise environment
• Development data was captured inside a moving car
• Cabin noise, engine sound, and road hum make recognition harder

Rubric: To receive full credit, the response must contrast the clean training conditions with the noisy vehicle recordings and explicitly state that the added car-related noise is the reason performance falls.