Algoritmos OPWI e LDM-GA para sistemas de conversão texto-fala de alta qualidade empregando a tecnologia SCAUS / Algorithm OPWI and LDM-GA for high quality text-to-speech synthesis based on automatic unit selection

AUTOR(ES)
DATA DE PUBLICAÇÃO

2006

RESUMO

This Thesis presents two new algorithms for Unit Selection Based Text-to-Speech systems (USBTTS). The first algorithm is the OPWI (Optimized Prototype Waveform Interpolation), which was designed to be used as a Back-End module for USB-TTS. The second algorithm is the LDM-GA (Linguistic Data Mining Using Genetic AIgorithm), which was designed to minimize training problems related to LNRE (Large Number of Rare Events) distributions. Experimental results and analysis of the OPWI and LDM-GA algorithms are presented in detail. The OPWI algorithm is evaluated under operations af analysisjre-synthesis and pr~sodic modifications, TSM (Time Scale Modifications) and PSM (Pitch Scale Modifications). The LDM-GA is evaluated in the context of phaneme segmental duration prediction based on linear regression mo de!. In addition to these two new algorithms (OPWI and LDM-GA), this Thesis presents a large review of the main modules of a USB-TTS system,Front-End Module (Linguistic module), prosodic module, unit-selection module and Back-End module (Synthesis module)

ASSUNTO(S)

inteligencia artificial linguistica - processamento de dados sistemas de telecomunicação processamento de sinais - tecnicas digitais digital signal processing processamento de sinais sistemas de processamento de fala machine learning techniques telecommunication systems computational linguistics

ACESSO AO ARTIGO

Documentos Relacionados