Multi-channel audio processing techniques

Abstract

The first issue of this thesis, new spatial audio coding schemes are proposed as a multichannel audio coding scheme. The SAC is a process to represent multichannel audio signals as down-mixed signal with spatial cues. Recently, binaural cue coding (BCC) has been introduced and becomes an important scheme for spatial audio coding. The inter-channel level difference (ICLD) as one of spatial cues of the BCC plays a pivotal role to remove a lot of redundant information. The accuracy of the ICLD, however, can be easily distorted by a quantization process. Instead of the ICLD, a new representation method of ICLD is proposed and it dramatically overcomes the quantization distortion. Another proposed scheme, global vector split based virtual source location information is newly presented as a SAC scheme. The GS-VSLI is analyzed on the semicircle plane and represented as angles. Spectral distortion measurement is conducted to confirm the usefulness of the GS-VSLI. As the second issue of the thesis, audio source separation techniques are dealt with. The object-based audio rendering is a method in order to make an auditory scene automatically. The core technique to realize object based audio processing is a blind source separation which makes multitude audio separated into object audio. For the robustness of our algorithm, the frequency-domain block-based multichannel blind deconvolution (MBD) with a normalization matrix is proposed. The normalization is designed to overcome the intrinsic problems of the time-domain MBD such as the whitening effect and the slow convergence. The experimental results confirm that the proposed MBD algorithm is superior to the previous works.