Research On Speaker Recognition System And Its Influence On Stuffy Nose

Voice is an inherent feature of human beings,and the cost of acquiring equipment is relatively low and is not limited by location.Therefore,speaker recognition technology that uses voice to identify speaker identity has many unique advantages and is even irreplaceable in many scenarios.However,the development of speaker recognition technology still faces many difficulties and challenges.Affected by the vocal organs,nasal congestion sounds when the nasal cavity is blocked will affect the recognition rate of the speaker recognition system.Therefore,this article will study the difference between nasal congestion sounds and normal sounds,analyze several speaker recognition systems and the extent to which they are affected by nasal congestion sounds,and provide guidance for selecting speaker recognition systems in practical applications.This article first analyzes the characteristics of the voice signal,extracts the system flow according to the voice feature,and introduces the digitization method of the voice;through the experiment,analyzes the necessity of the voice preprocessing process of pre-emphasis,framed windowing,and spectral entropy endpoint detection;The principle of extracting the characteristics of voiceprint by Mel cepstrum coefficients is introduced.Through the above operations,the original voice is converted into a digital signal that can be input into the system.Then,the mechanism of normal and nasal congestion sounds was studied,and the nasal congestion speech was analyzed by spectrogram technique to find out the specific difference between nasal congestion speech and normal speech.Then study the principle and performance of vector quantization method and Gaussian mixture model method in traditional speaker recognition system,find the best system parameters through experiments,compare the recognition rate of the best system with the effects of nasal congestion and analyze the experimental data.Finally,the principle and performance of the BP neural network method and the cyclic neural network algorithm based on integrated learning thought optimization in the neural network speaker recognition system are studied.The best system parameters are found through experiments,the best systems are affected by nasal congestion sounds,and the experimental data are analyzed.Through experiments,two types of four algorithms were realized,and the accuracy of single,double and double nasal congestion sounds at 4,8,16 and 24 persons were tested.Overall,the vector quantization method in traditional algorithms has a higher recognition rate of nasal congestion than the Gaussian mixture model algorithm,the training time of the Gaussian mixture model is short,and the recognition time of the vector quantization method is short.The neural network algorithm is less affected by the nasal congestion sound than the traditional algorithm,and the degree of impact does not change with the increase of the number of people.The training time and recognition time after optimization are shorter,but the neural network has to retrain all samples when increasing the number of people based on the characteristics of the system.The traditional algorithm only needs to train the increased samples.The selection method can be considered comprehensively in practical applications.