Research On Vehicle Sound Conversion System Based On MGC Parameter And BP Neural Network

With the wide application of automobiles,people pay more and more attention to the sound quality of automobiles.Most of the sound signal acquisition systems used in the world is Artificial HEAD testing system.However,due to the high price and confidentiality technology of this equipment system,the research of automobile sound quality has been restricted to varying degrees.The portable data acquisition system independently studied has been able to obtain accurate sound pressure level.However,it does not take into account the auditory characteristics of the human ear,it still does not meet the application requirements.This atticle establishes a voice conversion system based on MGC(Mel-Generalized Cepstrum)parameter and BP(Back Propagation)neural network and a voice evaluation model,which effectively solves this problem.Firstly,this article establishes the portable data acquisition system of sound signal by using ordinary microphone and the front-end adjustment equipment.Experiments are carried out in semi-anechoic chamber and on the wide road with two electric vehicles.At the same time,the portable and Artificial HEAD testing system are used to collect the vehicle sound signal under various working conditions.A total of 156 groups of sound sample data are obtained for post-processing.Through preliminary analysis of the original waveform and spectrum of the sound signal,it can be seen that there are some differences in frequency and amplitude between the two acquisition systems.Subsequently,a voice conversion system based on MGC parameter and BP neural network has been established.In this paper,Mel generalized cepstrum(MGC)parameter biased to human auditory characteristics is selected.Through comparison,this article chooses WORLD sound analysis and synthesis system,which has excellent effects of the synthesized sound.This system can extract three characteristic parameters of the sound signal through three algorithms of this system: fundamental frequency,spectrum envelope and aperiodic parameter.Then,a four-layer BP neural network conversion model based on MGC parameter has been established.This article synthesizes the new sound by taking the converted sound characteristic parameters into WORLD sound analysis and synthesis system and denoises the new sound by threelayer wavelet decomposition.By analyzing the original waveform and spectrum,it can be seen that the quality of the synthesized sound signal has been greatly improved.Finally,the sound evaluation model has been established.The error of the synthesized and target sound signals is evaluated by using the evaluation standard in the field of sound conversion and sound quality.The ABX and MOS tests show that the newly synthesized sound signal is more inclined to the target sound signal and the sound quality is better.By comparing the loudness,roughness,sharpness and sound level A in the field of sound quality evaluation,the average error of the new synthesized sound signal has met the engineering requirements.It shows that the sound conversion model and the sound evaluation model studied in this paper can effectively reduce the error of portable data acquisition system and achieve the purpose of engineering application.In this article,a complete sound conversion model and sound evaluation model from the point of view of sound signal processing have been established.This article converts the collected sound signal into the target sound signal collected by Artificial HEAD testing system with the sound conversion model,by extracting the characteristic parameters of the sound signal.Then,it establishes the sound evaluation model.After the error analysis,it can be preliminarily shown that Artificial HEAD testing system can be basically replaced in the evaluation of sound quality.