UniAudioCodecTrainingOptions

Options & Configuration · AiDotNet.TextToSpeech.CodecBased

The released universal codec’s adversarial-pretraining configuration.

Constructors

Constructor Summary
UniAudioCodecTrainingOptions() Creates an instance with the default settings.
UniAudioCodecTrainingOptions(UniAudioCodecTrainingOptions other) Copies the settings from another options instance.

Properties

Property Summary
double CommitmentWeight { get; set; } Gets or sets the weight of the commitment objective.
double EpochDecay { get; set; } The released exponential learning-rate multiplier per completed codec epoch.
double FeatureMatchingWeight { get; set; } Gets or sets the weight of the feature matching objective.
double FullStftWeight { get; set; } Gets or sets the weight of the full STFT objective.
double LearningRate { get; set; } AdamW rate for both codec and critic; betas are (0.8, 0.99), epsilon 1e-6.
double SubStftWeight { get; set; } Gets or sets the weight of the sub STFT objective.
double WeightDecay { get; set; } Gets or sets the optimizer weight decay.
double[] Bandwidths { get; set; } Training bandwidths in kbps; v1’s eight-stage codec uses the first five values.
int Epoch { get; set; } Completed codec epochs, retained with the training configuration when resuming.
int SegmentSize { get; set; } Samples in the released training segment (32,000).
int Subbands { get; set; } PQMF subbands for the sub-band spectral objective (six).
int[] DiscriminatorChannels { get; set; } Each critic’s final hidden width (paper Appendix E).
int[] DiscriminatorHops { get; set; } Critic STFT hop lengths; each FFT/window is four times its hop.
int[] FullFftSizes { get; set; } Gets or sets the full FFT sizes.
int[] FullHopSizes { get; set; } Gets or sets the full hop sizes.
int[] FullWindowSizes { get; set; } Gets or sets the full window sizes.
int[] SubFftSizes { get; set; } Gets or sets the sub FFT sizes.
int[] SubHopSizes { get; set; } Gets or sets the sub hop sizes.
int[] SubWindowSizes { get; set; } Gets or sets the sub window sizes.
long AdversarialStartStep { get; set; } The generator includes adversarial losses after this step (zero in the release).
long CompletedSteps { get; set; } Completed generator/discriminator step pairs.