Comparative analysis of natural and synthesized Polish speech

Abstract

In the evolving field of speech synthesis, not only intelligibility, but also naturalness remains an important factor. This paper presents a comparative analysis of natural versus synthesized Polish speech. Speech synthesizers: Ivona, Mekatron, Notevibes, and ttsmp3 were explored. Four methods for assessing synthesized speech quality and comparing it to natural speech were presented: the AB test, MOS, logatom articulation test, and MUSHRA. Sentence databases and a database of logatoms were generated for each synthesizer and recorded for natural speech. Results indicated natural speech was consistently better than synthesized speech. Among the synthesizers, Notevibes performed best in all comparisons, while Mekatron ranked lowest.

Keywords

• synthesized speech
• ab test
• mos
• mushra
• logatom articulation