AlignTTSTrainingPhase

Enums · AiDotNet.TextToSpeech.Classic

The training phases of AlignTTS (Zeng et al. 2020, §3.2.2), run in order.

Fields

Field Summary
AlignTTSTrainingPhase Alignment Train the mix density network and the character-side blocks with the alignment loss.
AlignTTSTrainingPhase Decoder Freeze the character side; train the mel side on mel MSE with the MDN’s Viterbi durations.
AlignTTSTrainingPhase DurationPredictor Train the duration predictor on the final MDN durations (log-domain MSE).
AlignTTSTrainingPhase JointFineTuning Fine-tune the feed-forward Transformer and the MDN together, durations recomputed every step.
int value__