EUROSPEECH 2003 - INTERSPEECH 2003
8th European Conference on Speech Communication and Technology

Geneva, Switzerland
September 1-4, 2003

        

Evolutionary Weight Tuning Based on Diphone Pairs for Unit Selection Speech Synthesis

Francesc Alias (1), Xavier Llora (2)

(1) Ramon Llull University, Spain
(2) University of Illinois at Urbana-Champaign, USA

Unit selection text-to-speech (TTS) conversion is an ongoing research for the speech synthesis community. This paper is focused on tuning the weights involved in the target and concatenation cost metrics. We propose a method for automatically adjusting these weights simultaneously by means of diphone and triphone pairs. This method is based on techniques provided by the evolutionary computation community, taking advantage of their robustness in noisy domains. The experiments and their analyses demonstrate its good performance in this problem, thus, overcoming some constraints assumed by previous works and leading to a new interesting framework for further investigations. La conversio text-parla (CTP) basada en seleccio d'unitats es una de les linies de recerca actuals de la comunitat cientifica de sintesi de veu. Aquest treball se centra en l'ajust dels pesos involucrats en el calcul dels costos d'unitat i de concatenacio. Es presenta un metode automatic per l'ajust simultani d'aquests pesos a partir de parelles de difonemes i trifonemes. Aquest metode esta basat en tecniques obtingudes de la comunitat de computacio evolutiva, aprofitant la robustesa d'aquests algorismes en dominis sorollosos. Els experiments que s'han dut a terme i la seva posterior analisi demostren el bon funcionament del metode en aquest problema, ja que supera algunes de les restriccions d'anteriors metodes. A mes, esdeve un marc de treball molt interessant per a properes investigacions.

Full Paper

Bibliographic reference.  Alias, Francesc / Llora, Xavier (2003): "Evolutionary weight tuning based on diphone pairs for unit selection speech synthesis", In EUROSPEECH-2003, 1333-1336.