AdaSpeech2TrainingStep
Enums · AiDotNet.TextToSpeech.Classic
The steps of AdaSpeech 2’s adaptation pipeline (Yan et al. 2021, §2, Fig. 2) that train the model.
Fields
| Field | Summary |
|---|---|
AdaSpeech2TrainingStep MelEncoderAligning |
Mel encoder aligning: only the mel encoder trains, toward the expanded phoneme hidden sequence. |
AdaSpeech2TrainingStep SourceModel |
Source model training: AdaSpeech training on transcribed data. |
AdaSpeech2TrainingStep UntranscribedAdaptation |
Untranscribed speech adaptation: reconstruction through the mel encoder and decoder, adapting only the conditional layer normalizations and the speaker embedding. |
int value__ |