Table 2 Summary of supervised ETTS approaches. “LingF” stand for linguistic features, “EmoL” stands for emotion labels, “MelS” stands for mel-spectrograms, “PhnSq” stands for phoneme sequence, “ChrSq” stands for character sequence, “ProsF” stands for prosodic features, “LM-F” stands for features from a language model and “ET” stands for expression/emotion transplantation
Ref No. | Inputs | Emotion label representation | ET | TTS model |
|---|---|---|---|---|
[80] | LingF+EmoL | One-hot vector | DL-SPSS, HMM | |
[65] | LingF+EmoL | One-hot vector/dependent layers | \(\checkmark\) | DL-SPSS |
[66] | LingF+EmoL | Perception vector/matrix | DL-SPSS | |
[41] | LingF+EmoL | One-hot vector | DL-SPSS | |
[42] | LingF+EmoL | Dependent layers | DL-SPSS | |
[81] | LingF+EmoL | One-hot vector/set of neurons | \(\checkmark\) | DL-SPSS |
[43] | LingF+EmoL | One-hot vector/dependent layers/separated Model | DL-SPSS | |
[82] | LingF+EmoL | One-hot vector | \(\checkmark\) | DL-SPSS |
[83] | PhnSq+LM-F+EmoL | Embedding vector | Encode-Dttention-Decoder | |
LingF+EmoL | One-hot vector/dependent layers/separated Model | \(\checkmark\) | DL-SPSS | |
[26] | PhnSq+MelS+EmoL | One-hot vector as ground truth GSTs weights | Tacotron2 | |
[27] | PhnSq+LingF+EmoL | Embedding vector | Tacotron2 | |
[84] | LingF+EmoL | Joint embedding with other data labels | DL-SPSS | |
[85] | LingF+ProsF+EmoL | Ground truth for a classifier | DL-SPSS | |
[86] | PhnSq+EmoL | Embedding vector | Transformer TTS | |
ChrSq+MelS+EmoL | Ground truth for a classifier | Tacotron2 | ||
[69] | LingF+EmoL | One-hot vector/dependent layers | \(\checkmark\) | DL-SPSS |
[34] | PhnSq+MelS+EmoL | Ground truth for a classifier | Tacotron2 | |
[64] | ChrSq+LM-F+EmoL | Ground truth for a predictor | Tacotron2 | |
PhnSq+MelS+EmoL | Ground truth for a classifier | Tacotron2 |