Mini-Omni: Language Models Can Hear, Talk While Thinking in Streaming