Research And Implementation Of Audio Public Opinion Analysis System Based On Heterogeneous Neural Network

With the rapid development of mobile Internet technology,the public can freely express their opinions through the network,forward and comment on social hot events,which leads to the outbreak of network public opinion events.In recent years,the main body of public opinion event information has gradually changed from structured text data to audio-visual data with a higher degree of audio-visual.However,due to the difference in the duration of audio-visual data and the possibility of super long time,it is impossible to recognize and analyze this kind of audio data directly.At the same time,the existing speech recognition system does not integrate public opinion domain knowledge and audio emotional factors,It is unable to conduct effective public opinion research and judgment on audio and video data.To solve the above problems,this paper designs and implements an audio endpoint detection algorithm based on multi feature input,which realizes the segmentation of super long single file audio data according to semantic interval,and the short audio after segmentation can be analyzed and recognized directly.An audio semantic recognition model combined with intelligent error correction is designed and implemented,which transforms unstructured audio data into structured text description,and automatically corrects the recognition errors in the recognition results.Design and implement a public opinion research and judgment model integrating audio emotion,build a public opinion classification system combined with public opinion domain knowledge,and conduct audio public opinion research and judgment with reference to public opinion classification system and audio emotion factors.Finally,the concept of heterogeneous neural network is proposed,and multiple models are integrated to design and implement an audio public opinion analysis system based on heterogeneous neural network,which provides complete functions from audio and video data collection,data distributed transmission to audio public opinion analysis and results visualization.The experiment obtains the audio and video data of network community and open source voice database through the system acquisition module,and verifies the audio endpoint detection algorithm,audio semantic recognition model,audio emotion recognition model and public opinion judgment model respectively.Compared with the baseline model RCNN,the word error rate of the proposed audio semantic recognition model is reduced by 5.22%;compared with the 3d-acrnn model,the unweighted accuracy of the audio emotion recognition model is improved by 4.06%;the public opinion index calculation model and fasttext public opinion research and judgment model proposed in this paper are consistent with the specific public opinion Compared with the method of emotional event detection,the F value were increased.The experimental results show that the audio public opinion system based on heterogeneous neural network can effectively analyze the audio and video data in the network.