Figure 7.
Two stacked bar chart sections present Naturalness M O S and Quality M O S. Scores range from 1 to 5. Each section includes G T, speakers m 01 to m 05, speaker f 01 and the whole group. For each speaker and the whole group, T F, Ours and S A 2 are compared using score distributions and model means plus or minus 95 per cent confidence intervals. Naturalness M O S whole-group means are 2.77 plus or minus 0.05 for T F, 3.28 plus or minus 0.05 for Ours and 3.50 plus or minus 0.04 for S A 2. Quality M O S whole-group means are 2.62 plus or minus 0.05 for T F, 3.19 plus or minus 0.04 for Ours and 3.47 plus or minus 0.04 for S A 2. G T has means of 3.70 plus or minus 0.11 for naturalness and 3.70 plus or minus 0.10 for quality.Score distributions in subjective evaluations on the naturalness and sound quality of the synthesized explosion sounds (top: Naturalness MOS, bottom: Quality MOS). For m-01, the subjective evaluation results for Ground Truth are shown above. For each speaker, the frequency distribution of subjective scores is presented as a stacked bar chart. The corresponding MOS (mean ± 95% confidence interval) is displayed next to each bar