The final goal of this study is to establish a speech production model which can manipulate acoustic characteristics corresponding to place and manner of articulation, whilst maintaining aspects of the signal relating to speaker identity, and with the high signal quality required for speech synthesis.
The project consists of three parts: 1) resolving the conventional problem of estimating the transfer characteristics of the vocal tract; 2) finding a method suitable for altering the vocal tract response in order to manipulate articulation; 3) identifying a mapping between the transfer characteristic and an articulatory representation.
In this talk, I will be discussing the first part of this project. After pointing out several problems of conventional analysis/synthesis techniques that are based on source-filter models, I will explain the method which I have proposed to overcome those problems.