Single Channel Phase-Aware Signal Processing in Speech Communication

Chapter 5Phase Processing for Single-Channel Source Separation

Pejman Mowlaee and Florian Mayer

Graz University of Technology, Graz, Austria

5.1 Chapter Organization

In this chapter, we provide another practical application to demonstrate the importance of phase processing in speech communication. In contrast to Chapter 4, which is focused on speech enhancement, this chapter will examine the role of phase in single-channel source separation (SCSS). After a review of conventional phase-unaware SCSS methods, we proceed with several examples where phase processing has been incorporated for SCSS. We highlight the contribution of phase processing in two steps: spectral modification and signal reconstruction. The recent advances with regard to incorporating phase in SCSS for the signal interaction function, phase estimation for signal reconstruction, time–frequency masking, and matrix factorization for spectral modification will be thoroughly demonstrated. The chapter supplies various examples to visualize and quantify how phase processing pushes the achievable performance of the state-of-the-art phase-unaware SCSS methods.

5.2 Why Single-Channel Source Separation?

5.2.1 Background

We often encounter situations where simultaneous acoustic events occur, among which we might be interested only in one particular signal. For instance, imagine a busy intersection with car noise as well as an acoustic alarm warning or, as an alternative scenario, a room crowded with people. Cherry termed ...