Challenges in Automated Processing of Speech from Child Wearables: The Case of Voice Type Classifier