SPEARTTSTrainingStage
Enums · AiDotNet.TextToSpeech.CodecBased
Which SPEAR-TTS network, and which of its training tasks, a training step runs.
Fields
| Field | Summary |
|---|---|
SPEARTTSTrainingStage Backtranslation |
The backtranslation model’s finetuning (§4.2): semantic tokens to text, updating the decoder’s top and bottom layers. |
SPEARTTSTrainingStage Pretraining |
S1’s pretraining (§4.1): reconstruct semantic tokens from a copy with tokens deleted at random. |
SPEARTTSTrainingStage Reading |
S1’s finetuning (§4.1, §7.2): text to semantic tokens, updating the encoder’s embeddings and lower layers. |
SPEARTTSTrainingStage Speaking |
S2 (§5): semantic to acoustic tokens, prompted by another window of the same utterance. |
int value__ |