A Study On The Method Of Speech Recognition In Vietnamese Tourism

With the continuous development of science and technology, robot has been applied to all aspects of production and life. Thus, it can replace humans to proceed cumbersome and high-risk tasks, and make people’s living quality improved continuously. As an important part of the human-computer interaction technology, the speech recognition get more and more attention of the researchers. At present, the Vietnamese speech recognition research focuses on the key technology such as acoustic model and language model. However, no matter acoustic level or language level, there are obvious differences between terms in different field. It makes the speech recognition method in general field difficult to apply to specific fields directly. In this paper, we aim to building a question speech recognition system in the area of Vietnamese tourist. We mainly discuss the acoustic model building method based on acoustic characteristics of domain term and the language model building method integrated the domain knowledge. The main work of this paper includes the following parts:(1)We study the acoustic model building method of Vietnamese questions in tourism areas. This method firstly determines the phonemes as recognition element, so that making the construction of acoustic model a moderate scale. And then, we analyze and statistic the distribution of the context for speech corpus of phonemes, and construct three phonemes acoustic model to solve the problem of Vietnamese co-articulation. Then, based on phoneme pronunciation characteristics, the decision tree problem set can be built. Using decision tree state sharing policy, we optimize the training process of acoustic model, so as to solve the data sparse problem. Finally, through the contrast test, we verify the effectiveness of building method of three phonemes acoustic model based on decision tree state share.(2)We study the language model building method of Vietnamese questions in tourism areas. We first determine language model category of questions speech recognition, which is suitable for tourism areas. Second, according to the differences of inquiry object in tourism areas, we divide the questions form. Then, the grammar rules of Vietnamese questions you need to follow are summarized. Finally, based on the statistics in different forms of questions structure, we determine the specific description of the language model, and build a language model of questions speech recognition, which is suitable for tourism areas.(3)Through the experiment, using different evaluation index of tourism field questions, the performance of speech recognition system are analyzed and summarized, and the effectiveness of this method is verified.