| In recent years,with the rapid development of social technology,especially the rapid development of artificial intelligence technology,convenient voice human-computer interaction technology has rapidly appeared in various application scenarios in the field of culture and sports.Most of the application scenarios of style and sports have the problem of excessive environmental noise,which seriously affects the reliability of the actual human-machine voice interaction.This type of problem is collectively referred to as the "cocktail party problem"(cocktail party problem),which has always been a hot spot in the field of speech recognition One of the research questions.Relying on the intelligent table tennis training project,this thesis first designed a set of intelligent table tennis training voice question and answer retrieval system for this problem.Considering that the application scenario of this system is the equipment of the table tennis club intelligent table tennis training system,it is in a noisy environment for playing table tennis.For this purpose,the voice processing part of the system is completed by the combination of three modules of voice noise reduction,voice separation and voiceprint recognition.The purpose is to remove environmental noise and other people’s voices,so that the final output is only the voice of the trainer.Secondly,the similarity calculation problem of intelligent question answering and retrieval system is discussed.The corpus of this thesis is based on the field of table tennis,and system questions are generally shorter sentences,so the focus of similarity calculation in this thesis is sentences.Based on the analysis of the principle of the correlation similarity algorithm,the thesis adds a position vector to the original vector based on the shortcomings of the traditional TF-IDF algorithm to achieve the purpose of performance improvement.In addition,this thesis also analyzes the similarity calculation method based on How Net.Aiming at the unclassified situation of semantic origin,this thesis proposes a similarity calculation method based on the actual corpus.Finally,two sets of control experiments are designed.The experimental results show that the improved similarity algorithm has better performance and meets the expected design requirements. |