Research On Unified Of Audio And Speech Coding Based On Compressed Sensing

With the wide application of portable multi-media equipments,the problem of sound signal's compression and coding with low bit rate and high fidelity has become one of the widely discussed topics for research scholars in the field of information processing still.In the recent thirty years,audio coding and speech coding technology have obtained fast development,but because of the type of sound signal's diversity,the existing speech coding and audio coding system can not provide a full transparent sound quality.Therefore,there is need a new compression coding technology for modern sound signal's processing system in order to code the speech signal and audio signal uniformly.Unified of speech and audio coding(USAC)technology has become a general topic for industry and researchers currently.At present,the MPEG-D USAC system is widely used,and it uses two independent core codec of speech and audio codec to process speech and audio signal respectively.MPEG-D USAC system classifies the audio and speech signal from the input signal by using the method of speech audio discrimination,and uses the enhanced spectral band replication(eSBR)technology to compress the high frequency components of the input signal.Different from the previous spectral band replication(SBR)technology,the eSBR module in USAC system needs to process speech and audio signal simultaneously,it uses multi-channel quadrature mirror filter(QMF)banks to realize the input sound signal's time-frequency transform,so leads to the USAC system with high complexity.Therefore,based on the MPEG-D USAC system in this paper,we use the spare fast Fourier transform(SFFT)technology to improve the eSBR module,and realize the design of low complexity scheme of MPEG-D USAC.In this paper,we use the SFFT sensing algorithm of low complexity to design the multi-channel QMF filter banks,and achieve signal's fast time-frequency transform.By using the SFFT transform to improve the eSBR module,it can be able to extract some important component of the signal in Fourier domain with sub linear time,and effectively reduce the computation that caused by signal's discrete Fourier transform.The experimental data show that,compared with the traditional eSBR technology,it can reduce the computational complexity by using the SFFT algorithm for signal's time-frequency transform and its running time can saves about several times.Finally,by evaluating and analyzing the signal's sound quality and wave that output from the improved US AC system,we can prove that the method proposed in this paper is greatly reduced the complexity of the USAC system,and can still efficiently encode and decode the input signal that mixed with speech and audio signal,so achieve the expected results of the theory.