DiTToCodecTrainingOptions

Options & Configuration · AiDotNet.TextToSpeech.FlowDiffusion

Separate Mel-VAE training, CLaM-TTS equations 4–6 and DiTTo equation 2.

How It Works

The papers do not disclose discriminator widths, crop lengths, resolution reductions or loss coefficients. These exposed defaults are implementation choices, not recovered private checkpoint settings. Semantic finetuning uses a frozen ByT5-base decoder and requires its pretrained weights.

Constructors

Constructor Summary
DiTToCodecTrainingOptions()
DiTToCodecTrainingOptions(DiTToCodecTrainingOptions other)

Properties

Property Summary
double AdversarialWeight { get; set; }
double CommitmentWeight { get; set; }
double FeatureMatchingWeight { get; set; }
double LearningRate { get; set; }
double ReconstructionWeight { get; set; }
double SemanticWeight { get; set; }
int DiscriminatorChannels { get; set; }
int SemanticDecoderLayers { get; set; }
int SemanticFeedForwardDim { get; set; }
int SemanticHeads { get; set; }
int SemanticKeyValueDim { get; set; }
int SemanticModelDim { get; set; }
int SemanticVocabulary { get; set; }
int[] DiscriminatorWindows { get; set; }
int[] ResolutionDivisors { get; set; }