Primary And Ambient Components Extraction For Audio Scene Reproduction

The suggestive primary component and the diffuse ambient component which create the atmosphere describe the whole content of the acoustic scene.For example,at the concert,the singer's voice is the primary component and the audience's cheers and applause is the ambient component.In the perception of the acoustic scene,the primary component provides cues,while the ambient component renders the atmosphere.The traditional audio processing in channel-based audio playback system ignores the difference in perception between primary and ambient components and blurs the distinction between them.In order to make the channel-based audio signal can be played on arbitrary playback system and pursue an immersive sound scene,the primary component and the ambient component need to be rendered by different ways.However,the existed channel-based audio signals can only provide mixed signals and the primary component and the ambient component need to be extracted.This extraction process is called the primary ambient component extraction(PAE).PAE acts as a front end promoting flexible,efficient,immersive spatial audio playback.Since PAE is an underdetermined problem essentially and depends on the specific signal model.The main purpose of this paper is to analyze and improve the PAE algorithm under the existing signal model as well as try to put forward a new signal model and its corresponding PAE method.The main work is as follows.This paper discusses the performance of the principal component analysis algorithm(PCA),least squares algorithm(LS),minimum distortion least squares algorithm(MDLS),minimum leakage least squares algorithm(MLLS)under a linear estimation framework as well as ambient phase estimation with a sparsity constraint algorithm(APES),rapid ambient phase estimation with a sparsity constraint algorithm(APEX)under a novel ambient spectrum estimation framework(ASE).The results show that the performance of the algorithms under ASE framework are better than those under a linear estimation framework.APES performs best but has a complex computation.Ambient phase estimation with non-uniform searching(APEN)algorithm is proposed in this paper.The performance of the algorithm is analyzed by objective evaluation and subjective evaluation.These two evaluation both show that APEN has very similar performance with APES and the computation complexity is much lower than APES.All above algorithms are applied in a time-shifted situation and the performance of them are discussed.This paper also proposes Random sign Hilbert filtering decorrelation process.This decorrelation process has more stable correlation coefficient and can effectively eliminate combing phenomenon.Corresponding to this decorrelation method,a new signal model and the uncorrelated-ambient PAE(UAPAE)are proposed in this paper.The results show that the UAPAE has highest extraction accuracy under this model.