What are the challenges and solutions for processing live audio streams with Cognitive Services?