Research On Voice Recognition In Security Monitoring

Human voice can well express their own emotions,so this paper studies how to use the emotions contained in human voice to monitor the safety of public places.The work is as follows:Considering that the public sound collected by the sound monitoring equipment will contain too much interference noise,the voice signal needs to be enhanced.In the traditional spectral subtraction method,the setting of the noise energy in the non-speech segment is the average energy of the noise,and the average energy is easily affected by the random impulse noise,which generally leads to the excessive average energy of the noise,and then leads to the loss of the speech signal spectrum in the spectral subtraction algorithm.In this paper,the traditional spectral subtraction speech enhancement algorithm is improved,and the noise energy ratio is introduced into the spectral subtraction algorithm.Experiments show that the improved spectral subtraction method not only reduces the risk of spectrum loss of speech signal,but also inhibits the noise in the speech signal.Considering that no research results have been made to clarify the emotions contained in public voices,this paper classifies the emotions contained in public voices into normal emotions,dangerous emotions,lively emotions and quiet and cheerless emotions.Collect emotional sounds from different scenes and build scene sound library.Combining the speech feature set and prosodic features of the international speech emotion recognition competition,four types of scene sound features are extracted.Considering the impact of penalty parameters and RBF kernel function width on the classification of c-svm model,this paper combines genetic algorithm and nonlinear programming to enhance the global and local search ability,so as to optimize the penalty parameters and kernel function width.Experimental results show that the random selection of penalty parameters and kernel width can only achieve 35% accuracy in the recognition of scene sound emotion.For the optimized selected parameters and,the recognition accuracy can reach 77%.Among them,the recognition rate of normal scene emotion and quiet scene emotion is the most stable and the highest,generally reaching 90%.The emotion of danger scene and the emotion of busy scene are easily confused,the recognition rate is between 60% and 80%,which affects the overall recognition rate of sound emotion in public places.