: Difference between revisions

Resumo

Palavras Chave: síntese de fala, sistemas texto-para-fala, texto-para-fala em português, modelo da excitação vozeada, modelo glotal, dinâmica dos parâmetros glotais

Abstract

The speech synthesis research currently faces two challenges: on one hand the improvement of the naturalness of synthetic speech and on the other the requirement to enlarge the number of languages covered by this technolo-gy in order to understand its specificities. This work intends to contribute to both questions.

In the first part we will focus our attention in the DIXI system, the first text-to-speech synthesis system built from scratch for Portuguese. We will present the system architecture compared with other systems developed for different languages, the design options and the development status of each system module, emphasizing the specific characteristics of the Portuguese language.

The naturalness of the synthetic speech depends on many specific factors of the language in question, but also on common aspects to other languages like the modeling of the human vocal apparatus. The second part of this thesis is dedicated to this problem, namely the modeling of the glottal air flow. We will describe the advantages and disadvantages of some of the available models, and present a new formulation that allows an automatic strategy to estimate its parameters from natural speech: the duration of the open glottis phase, the vocal folds closing speed and the amount of turbulence generated in the glottal flow.

Keywords: speech synthesis, text-to-speech systems, portuguese text-to-speech, voice source modeling, glottal model, glottal parameters dynamics

Full Text

• Síntese de Fala a Partir de Texto