Intelligent Speech Question Answering Algorithm And System Implementation Based On Convolutional Neural Network

2020 is an unusual year.The world has experienced an unprecedented new type of coronavirus COVID19.During the epidemic,many epidemic prevention measures have been introduced to curb the spread of the epidemic.As a result,the concept of "contactless" was put forward,and contactless human-computer interaction technologies such as intelligent voice question answering and interactive technologies have highlighted their advantages.Therefore,this article takes intelligent voice question answering as the research direction.The realization of voice question answering technology needs to solve three problems: voice recognition,the establishment of knowledge base,and answer matching.Speech recognition is the front-end data entry,and its recognition accuracy is directly related to the answer matching effect of the back-end question answering system.In this paper,the optimized convolutional neural network connection temporal(CNN-CTC)model is used to train and recognize the input speech;the establishment of the question and answer knowledge base requires rich content as data support.This article uses the Scrapy framework crawler to crawl a large amount of data on the web,segmentation and part-of-speech tagging of the data content,extracts entities based on heuristic rules and KNN algorithm screening and classification,uses PCNN algorithm to extract the relationships between entities,and combines entities,attributes and relationships In the form of triples,it is saved in the Neo4 j graph database to complete the creation of the knowledge base;the question and answer system uses the naive Bayes algorithm to match the question and the answer,and the similarity calculation method is used as a supplement to the question and answer.The two can complement each other to improve efficiency.The main work of this paper is as follows:(1)This article uses the convolutional neural network connection temporal(CNN-CTC)model,uses Mel-Frequency Cepstral Coefficient(MFCC)as the feature parameter,and continuously adjusts the CNN network structure through a large number of experiments to make the word error rate Reach the lowest,the model recognition rate reaches the best,can better recognize the user's voice problems.(2)With "medical" as the subject word,crawl a large number of medical-related vocabulary entity relationships,after data cleaning,store the data in the Neo4 j graph database to complete the construction of the knowledge graph.(3)Build a question and answer system,extract entities from user questions,classify user questions through KNN classification and filter,obtain user intentions,construct Cypher query sentences,query in the database for answers that match or similar to the questions and output,to achieve intelligent question and answer the goal of.