Experiments Towards the Creation of a Synthetic Dramatic Actor

Text to Speech technology has been evolving for the last 30 years and is finding use in a broad range of applications. This technology is attempting to apply modern Artificial Intelligence ideas to help it more closely resemble the capabilities of human speech. The speech community has also been expressing interest in applications such a synthetic dramatic reader. This paper aims to evaluate the potential and challenges for the creation of a synthetic dramatic actor/reader with the best of breed technology developed to date. It provides a quick historical review of the evolution of text to speech (TTS) technology, through present time, with consideration of the trade offs of the various technologies. It also provides a brief tutorial of the foundations for applying a new technology that uses statistics behind average voices and HMM synthesis to create synthetic speech adapted to a specific voice. The paper also describes the results of two experiments with modern TTS technology. The first experiment illustrates the ability of adaptive filter technology to vary the prosodic features that would to be expressed by a dramatic actor. The second experiment shows the capability of average voice HMM synthesis techniques to clone a voice. The paper ends with a discussion of the possible new applications of this technology, and an evaluation of its potential for creating a synthetic actor. It ends with a discussion of planned future work.