Insights On Real-World Extemporaneous Speech Systems