Application of sound source separation methods to advanced spatial audio systems

Abstract

This thesis is related to the field of Sound Source Separation (SSS). It addresses the development and evaluation of these techniques for their application in the resynthesis of high-realism sound scenes by means of Wave Field Synthesis (WFS). Because the vast majority of audio recordings are preserved in twochannel stereo format, special up-converters are required to use advanced spatial audio reproduction formats, such as WFS. This is due to the fact that WFS needs the original source signals to be available, in order to accurately synthesize the acoustic field inside an extended listening area. Thus, an object-based mixing is required. Source separation problems in digital signal processing are those in which several signals have been mixed together and the objective is to find out what the original signals were. Therefore, SSS algorithms can be applied to existing two-channel mixtures to extract the different objects that compose the stereo scene. Unfortunately, most stereo mixtures are underdetermined, i.e., there are more sound sources than audio channels. This condition makes the SSS problem especially difficult and stronger assumptions have to be taken, often related to the sparsity of the sources under some signal transformation. This thesis is focused on the application of SSS techniques to the spatial sound reproduction field. As a result, its contributions can be categorized within these two areas. First, two underdetermined SSS methods are proposed to deal efficiently with the separation of stereo sound mixtures. These techniques are based on a multi-level thresholding segmentation approach, which enables to perform a fast and unsupervised separation of sound sources in the time-frequency domain. Although both techniques rely on the same clustering type, the features considered by each of them are related to different localization cues that enable to perform separation of either instantaneous or real mixtures.Additionally, two post-processing techniques aimed at improving the isolation of the separated sources are proposed. The performance achieved by several SSS methods in the resynthesis of WFS sound scenes is afterwards evaluated by means of listening tests, paying special attention to the change observed in the perceived spatial attributes. Although the estimated sources are distorted versions of the original ones, the masking effects involved in their spatial remixing make artifacts less perceptible, which improves the overall assessed quality. Finally, some novel developments related to the application of time-frequency processing to source localization and enhanced sound reproduction are presented.