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:

Much better +3
Better +2
Slightly better +1
About the same 0
  • 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

Pair File1 +3 +2 +1 0 +1 +2 +3 File2
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15

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.