With the continuous development of artificial intelligence industry, the technology of question-answering intelligent voice has been widely concerned and studied by scholars at home and abroad. However, there are two technical bottlenecks in speech recognition. The first lies in the speech recognition system, and the second lies in answering to the recognized voice. In response, this paper develops a question answering system with intelligent voice based on in-depth learning. This paper introduces the speech recognition system based on Hidden Markov Model at first , then studies the extraction technology of speech signal features based on Meier frequency, establishes some acoustic and linguistic models, and finally studies the question-answer matching model based on GRU algorithm. A question answering system has been developed as a result of above study. The experimental results show that the proposed algorithm, with considerable practical value, has higher accuracy in speech recognition and question answering than the traditional algorithm.