EC Application in Speech Processing - Voice Quality Conversion Using Interactive Evolution of Prosodic Control

Abstract

This chapter outlines Interactive Evolutionary Computation (IEC) and IEC application in speech processing. IEC is an optimization method that adapts evolutionary computation (EC) based on subjectively human evaluation. EC is a biologically inspired general computational algorithm and includes genetic algorithms (GA), genetic programming (GP), evolution strategies (ES), and evolutionary programming (EP) (Back et al., 1997). These algorithms are a heuristic search technique based on the ideas of natural selection and work with a population of solutions that undergo modifications under the influence of genetic manipulations and finally converge to the optimal solution. Here, the selection will be processed based on an explicit evaluative function prepared by human designer beforehand. EC has been widely used to engineering applications and effective for a variety of complex large sized combinational problems such as the travelling salesman problem, the job-shop scheduling problem, the machine-loading problem, etc. On the other hand, it is difficult, or even impossible, to design human evaluation explicit functions for interactive systems. For example, to obtain the most favorable outputs from interactive systems that create or retrieve graphics or music, such outputs must be subjectively evaluated. IEC is the technology that EC optimizes the target systems based on subjective human evaluation as fitness values for system outputs and effective for such kind of interactive systems. IEC research conducted during 1990s originated from the work of Dawkins (Dawkins, 1986) and two major research streams developed during the 1990s. The first research stream (Sims, 1991; Biles, 1994; Unemi, 2002) is Artificial Life (Dawkins, 1989) and the second research stream comes from the increase of researchers who are interested in humanized technology or human-related systems and have applied the IEC to engineering fields (Watanabe & Takagi, 1995; Takagi & Ohya, 1996; Sato, 1997; Parmee & Bonham, 1999; Kim & Cho, 2000). In the following, as a concrete example, we describe about voice quality conversion using IEC. In section 2 we briefly explain about a basic voice conversion technique and application. In section 3 we show voice elements and our voice quality conversion systems. In section 4 we describe prosodic coefficient fitting by IEC. In section 5 we describe the simulations we performed to evaluate this technique, and in the final sections we discuss our results and finish with a conclusion.