Goal of this Preference test

The goal is to evaluate the quality of the speech produce by an HMM-based synthesis system for French using a new glottal excitation model.

Instructions

1. For each line of the tab, listen carefully to file1 and file2 (but try to avoid to repeat the playing).

2. Then give a preference score about the timbre of the voice (please don't take the intonation into account), according to the following grades tab:

• to the left if you prefer file1
• to the right if you prefer file2.

3. When you answered all questions, verify all questions got an answer

4. Click on the button "Send the answers"

The test

Results

(based on 48 French naive listeners )

Box plots for each pair of audiofiles:

Ranking of the CCR test for the following speech samples: STRAIGHT-based system, SVLN-based system, Pulse train excitation-based system. The 95\% confidence intervals are presented for each score.