Influence of a voice on the quality of synthesized speech

Author

Hinterleitner, Florian ; Manolaina, Christiana ; Moller, Sebastian

Author_Institution

Quality & Usability Lab., Tech. Univ. Berlin, Berlin, Germany

fYear

2014

fDate

18-20 Sept. 2014

Firstpage

99

Lastpage

104

Abstract

In this study we focus on the influence of the voice of a speech corpus on the quality of text-to-speech (TTS) systems. Therefore, we selected a set of attribute scales which consists of items related to the 5 perceptual quality dimensions of TTS systems and of items which are used in studies concerning the likeability of voices. We selected 5 different voices of the TTS systems IVONA and ACAPELA and evaluated them in a listening test. A subsequent factor analysis revealed 4 factors. The first two factors consist of items related to the 5 perceptual quality dimensions whereas factor 3 and 4 are related to items which are used for likeability assessment. In a statistical analysis we could prove a significant effect of the voice of a speech corpus on the overall impression as well as on the factors 1, 2, and 4.

Keywords

speech synthesis; statistical analysis; ACAPELA; IVONA; TTS systems; attribute scales; listening test; perceptual quality dimensions; statistical analysis; subsequent factor analysis; synthesized speech quality; text-to-speech system quality; voice likeability assessment; Conferences; Correlation; Materials; Multimedia communication; Predictive models; Speech; Synthesizers; perceptual quality dimensions; quality prediction; speech corpus; text-to-speech; voice;

fLanguage

English

Publisher

ieee

Conference_Titel

Quality of Multimedia Experience (QoMEX), 2014 Sixth International Workshop on

Conference_Location

Singapore

Type

conf

DOI

10.1109/QoMEX.2014.6982303

Filename

6982303