Primary-Ambient Seperation Of Audio Signals / Karim Magdi Abdelfattah Ibrahim

Supervisor: Mahmoud Allam Key words:
Audio Source Separation, Primary/ambient Separation, Spatial Audio, Surrounding Sound
Systems, Upmixing.

Thesis (M.A.)—Nile University, Egypt, 2016 .

"Includes bibliographical references"

Contents:
1 Introduction 1
1.1 Blind Audio Source Separation (BASS) . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Primary-Ambient Extraction (PAE) . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2.1 Signal model for PAE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2.2 Sound localization and human auditory system . . . . . . . . . . . . . . . 6
1.3 Applications of PAE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.4 PAE assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.5 Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2 RelatedWork 11
2.1 Principal Component Analysis (PCA) . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.2 Inter-channel CoherenceMethod . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.3 Mid-side Decomposition Approach . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3 ProposedMethods 19
3.1 Improving The Principal Component Analysis (PCA)Method . . . . . . . . . . . 19
3.2 Neural NetworkMethod . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
4 Evaluation 29
4.1 Objective Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
4.1.1 BASS based evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
4.1.2 Single-source-extraction evaluation . . . . . . . . . . . . . . . . . . . . . . 34
4.2 Subjective Evaluation

Abstract:
Most audio recordings are in the formof a 2-channel stereo recording. However,many of the
new playback sound systems are designed to give a more spatial and surrounding atmosphere
that is beyond the content of the stereo recording. Hence, it is an essential step to extract
more spatial information from stereo recording in order to reach an enhanced upmixing
techniques. One approach of solving the problem is extracting the primary/ambient sources.
The problem of primary-ambient extraction (PAE) is one of the challenging problems to
decompose a signal into a primary (direct) and ambient (surrounding) sources based on their
spatial features. Several approaches have been used to solve the problem, however, there
is no clear way to evaluate the different methods of PAE. In this thesis, we propose a new
approach to solve the problem based on using trained neural networks aswell as modifications
to previous methods based on Principal Component Analysis (PCA) to improve their accuracy.
An evaluation method is also introduced to provide a way of comparison between different
methods. An objective and subjective evaluation is performed to compare between the current
state-of-the-art methods and our new proposedmethods.

Text in English, abstracts in English.