Skip to main content

Table 15 RMSEs for the four prosodic–acoustic features

From: Punctuation-generation-inspired linguistic features for Mandarin prosody generation

Feature set combinations lf0 (logHz) Dur (ms) Eng (dB) Pau (ms)
BSL BSL1 = Raw + G2P .191 43.77 3.72 71.73
BSL2 = BSL1 + WordSeg .182 39.93 3.53 64.62
BSL3 = BSL2 + WordPos .186 39.23 3.50 59.56
PCset PC1 = BSL3 + bPC .185 38.33 3.48 58.29
PC2 = BSL3 + iPCst .175 37.82 3.43 57.29
PC3 = BSL3 + iPCef .174 37.34 3.47 58.72
PC4 = BSL2 + iPCst .173 38.39 3.46 63.93
PC5 = BSL2 + iPCef .174 38.05 3.48 62.56
QCset QC1 = BSL3 + bQC .170 37.70 3.52 58.66
QC2 = BSL3 + sQC .169 37.83 3.52 57.95
QC3 = BSL2 + bQC .176 39.83 3.44 64.50
QC4 = BSL2 + sQC .172 39.30 3.54 63.33