Artificial Reverberation Method Based On Delay Network And Convolution Method

Indoor reverberation is a phenomenon in which sound is gradually attenuated after being emitted from the sound source due to continuous reflection and absorption by the indoor surface.If the reverberation is too large,the sound will become turbid and cluttered,and dereverberation technology is required.For example,dereverberation is used in a hands-free communication system to improve speech intelligibility and accuracy of speech recognition.However,if there is no proper reverb,the sound or music will sound very "dry" and uncomfortable to listen to.At this time,artificial reverberation technology can be used to shape the spatial sense of the sound and make the sound sound more natural and real.Therefore,artificial reverberation is widely used in music,movies,and virtual reality to beautify the timbre of music and sound,recreate art,and produce special sound effects.With the rapid development of science and technology in recent years,artificial reverberation technology is also closely related to more and more fields,and the research of artificial reverberation algorithms has also received more and more attention in the field of speech signal processing.This article mainly researches and improves the artificial reverberation method from the following two aspects:(1)Improve the two existing feedback delay network(FDN)methods,that is,the FDN algorithm proposed by Smith and the method proposed by Professor Bullen(which has been applied to Apple's application Sound Soup).The method proposed in this paper is based on the Householder feedback matrix and Butterworth filter bank.The number of delay lines is relatively free to choose,and the subband reverberation time can be accurately determined.In simulation experiments,the effects of the three methods are compared by using spectrum maps,speech quality perception evaluation,and listening experiments,which proves the effectiveness of the proposed method.(2)Investigate a convolution artificial reverberation method to decompose the room impulse response(RIR)into two parts: early reverberation and late reverberation: Early reverberation can be expressed as a sum of delays with corresponding attenuation factors;The late reverberation RIR is synthesized in the frequency domain and then transformed with the time domain.First,the room frequency response(RFR)is synthesized in the frequency domain using an autoregressive moving average model,and then obtained by inverse Fourier transform.Time-domain RIR.Finally,the clean and no-reverb signal is convolved with the RIR to obtain the reverb signal.In the simulation experiments,the effects of the three algorithms are compared using speech quality perception evaluation and listening experiments,which proves the effectiveness of the convolution method in this paper.