Figure 5.
Stacked bars present score distributions from 1 to 5 for speakers m 01 to m 05, speaker f 01 and the whole group. Each speaker compares T F, Ours and S A 2. Model means are reported with plus or minus 95 per cent confidence intervals. For the whole group, T F has 1.44 plus or minus 0.03, Ours has 2.43 plus or minus 0.05, and S A 2 has 2.69 plus or minus 0.05. Across individual speakers, T F means range from 1.29 to 1.57. Ours ranges from 2.22 to 2.67. S A 2 ranges from 2.52 to 2.89.Score distributions in subjective evaluations of fidelity. For each speaker, the frequency distribution of similarity scores between the reference sounds and the synthesized sounds from each model is shown as a stacked bar chart. The corresponding Fidelity MOS (mean ± 95% confidence interval) is displayed next to each bar