AlignTTSOptions

Options & Configuration · AiDotNet.TextToSpeech.Classic

Options for AlignTTS (feed-forward TTS that learns its own alignment with a mix density network).

Inheritance: ModelOptions → TtsModelOptions → AcousticModelOptions → AlignTTSOptions

For Beginners

These options configure the AlignTTS model. Default values follow the original paper settings.

How It Works

Defaults are the paper’s configuration (Zeng et al. 2020, §4.2): 6 FFT blocks on each side with every dimension 768, 2 heads and kernel 3; a duration predictor of 2 FFT blocks with dimension 128; a mix density network with hidden size 256 whose output is a mean and a variance per mel channel.

LearningRate configures the fixed rate in joint fine-tuning and duration-predictor training. Alignment and decoder pre-training retain the dimension-scaled Noam schedule adopted by the paper.

Constructors

Constructor Summary
AlignTTSOptions() Initializes a new instance with default values.
AlignTTSOptions(AlignTTSOptions other) Initializes a new instance by copying from another instance.

Properties

Property Summary
int DecoderDim { get; set; } Mel projection width, independent of the FFT hidden width (Zeng et al., Fig.
int DurationPredictorDim { get; set; } Gets or sets the duration predictor’s dimension (128).
int DurationPredictorLayers { get; set; } Gets or sets the duration predictor’s FFT blocks (2).
int FftFilterSize { get; set; } Gets or sets the FFT blocks’ convolution width (“the dimension of each network …
int FftKernelSize { get; set; } Gets or sets the kernel of both FFT convolutions (3).
int MixDensityHiddenLayers { get; set; } Gets or sets the mix density network’s hidden linear layers.
int MixDensityHiddenSize { get; set; } Gets or sets the mix density network’s hidden size (256).
int WarmupSteps { get; set; } Gets or sets the warmup steps of the first two stages’ inverse-square-root (Noam) schedule.