Speech Quality Test Results: MUSHRA naturalness scores with 95% confidence interval and Word Error Rate (WER). The column “GT or Pred” indicates whether we use ground-truth hierarchical ED (“GT”) or a text-predicted version. In the “TTS Model” column, “VA” and “VA(Multi-Step)” denote the TTS models employing the Single-Step hierarchical ED Variance Adaptor (Figure 4(d)) and the Multi-Step hierarchical ED Variance Adaptor (Figure 4(b)), respectively. Finally, the “Pred Mode” column specifies whether we predict the hierarchical ED sequentially from longer to shorter segments (Multi-Step) or in parallel for all segments (Single-Step)