FastSpeechOptions
Options & Configuration · AiDotNet.TextToSpeech.Classic
Options for FastSpeech (non-autoregressive TTS with duration predictor).
Inheritance: ModelOptions → TtsModelOptions → AcousticModelOptions → FastSpeechOptions
For Beginners
These options configure the FastSpeech model. Default values follow the original paper settings.
How It Works
Defaults are the paper’s LJSpeech configuration (Ren et al. 2019, §4.2 and Table 5): 6 FFT blocks on each side, hidden 384, 2 heads, convolutional FFN 384→1536→384 with kernel 3 in both layers, duration predictor of 256 filters and kernel 3, dropout 0.1.
Constructors
| Constructor | Summary |
|---|---|
FastSpeechOptions() |
Initializes a new instance with default values. |
FastSpeechOptions(FastSpeechOptions other) |
Initializes a new instance by copying from another instance. |
Properties
| Property | Summary |
|---|---|
double DurationPredictorDropout { get; set; } |
Gets or sets the duration predictor dropout (the paper’s dropout, 0.1). |
double DurationScale { get; set; } |
Gets or sets the speed control α applied to predicted durations at inference (Ren et al. |
int DecoderDim { get; set; } |
Decoder output width in mel channels (§4.2), independent of the FFT hidden width. |
int DurationPredictorFilterSize { get; set; } |
Gets or sets the duration predictor filter size. |
int DurationPredictorKernelSize { get; set; } |
Gets or sets the duration predictor kernel size (3). |
int FftFilterSize { get; set; } |
Gets or sets the filter size of the FFT blocks’ first convolution (1536). |
int[] FftKernelSizes { get; set; } |
Gets or sets the kernel sizes of the FFT blocks’ two convolutions (3 and 3). |