We've improved the Speaker Label model’s ability to identify unique speakers for single word or short utterances.