Hidden Markov Model: theorical recall & applications Claudio Zito ABSTRACT Although iniatially introduced and studied in the late 1960s and early 1970s, statistical methods of Markov source or Hidden Markov modelling have become increasingly popular in the last two decades. There are two strong reasons why this has occured. First the models are very rich of in mathematical structure and hence can form the theorical basis for use in a wide range of applications. Second, the models, when applicated proprerly, work very well in practice for several importatn applications. In the first part of this paper we attempt to review theorical aspects of this type of statistical modelling, focusing on the three basic problems for Hidden Markov Model. We present the solution to each of the three fundamental problems which it has been adopted in speech synthesis context. In the second part of this paper we show a short introduction to the branch of speech synthesis. We shortly cover its history from first approches until the current state of the art, focusing on the HMM-based speech synthesis system in which speech waveform is generated from HMMs themself. The relation between the HMM-based approch and other unit selection approches is also discussed.
REFERENCES L.R. Rabinier. A tutorial on Hidden Markov Models and Selected Applications in Speech Recognition. In Proceeding IEEE, 77(2):257285, 1989 K. Tokuda, H. Zen, A.W. Black, AN HMM-BASED SPEECH SYNTHESIS SYSTEM APPLIED TO ENGLISH, 2002 History and Development of Speech Synthesis, url: http://www.acoustics.hut.fi/publications/files/theses/lemmetty_mst/chap2. html HMM-based Speech Synthesis System (HTS), url: http://hts.sp.nitech.ac.jp/
Effective Multilingua Interaction in Mobile Environments (EMIME), url: http://www.emime.org/learn/speech-synthesis/read/tutorial-slides-ofhmm-based-speech-synthesis