Évaluation segmentale du système de synthèse HTS pour le français (Segmental evaluation of HTS) [in French]

Segmental evaluation of HTS HTS is a parametric speech synthesis system based on the use of Hidden Markov Models (HMM). HTS is now widely used but very few studies have been conducted to analyze the influence of parameters on the quality of the synthetic speech. The aim of this paper is to provide an objective evaluation of the quality of the synthetic speech produced by HTS in order to assess the influence of multiple descriptions. Our study concerns only the speech spectrum analysis and is applied to French. We propose to use GMM to measure the degradations introduced by HTS compared to reference voice. Finally, we propose a listening test in order to calibrate our objective measure. This method indicates that using other descriptors than the previous-current-next phoneme sequence does not improve significantly the modelisation of the spectrum. MOTS-CLÉS : HTS, qualité segmentale, évaluation, GMM.