SpeedySpeechOptions
Options & Configuration · AiDotNet.TextToSpeech.Classic
Options for SpeedySpeech’s student network (fully convolutional non-autoregressive TTS).
Inheritance: ModelOptions → TtsModelOptions → AcousticModelOptions → SpeedySpeechOptions
For Beginners
These options configure the SpeedySpeech model. Default values follow the original paper settings.
How It Works
Defaults are the paper’s (Vainer & Dušek 2020, §4.3) with what it leaves unstated taken from the authors’
implementation (github.com/janvainer/speedyspeech, hparam.HPFertility, HPStft): 128 channels; 13 encoder
residual blocks of two convolutions with dilations 4 × [1, 2, 4] + [1] (the paper’s “26 encoder blocks with
dilations 1, 1, 2, 2, 4, 4”), 17 decoder blocks with 4 × [1, 2, 4, 8] + [1] (its “34 decoder blocks”), kernel 4,
a duration predictor of three one-convolution blocks with kernels 4, 3 and 1 (the paper’s “dilations 4, 3, 1”);
LJSpeech audio at 22.05 kHz, 1024-point STFT, hop 256, 80 mel bins; log-mel targets standardized with the LJSpeech
statistics (mean −5.522, standard deviation 2.063); Adam at 0.002 with gradients clipped to norm 1.
Constructors
| Constructor | Summary |
|---|---|
SpeedySpeechOptions() |
Creates the paper’s LJSpeech configuration. |
SpeedySpeechOptions(SpeedySpeechOptions other) |
Initializes a new instance by copying from another instance. |
Properties
| Property | Summary |
|---|---|
double GradientClipNorm { get; set; } |
Gets or sets the gradient-norm clip (1). |
double MelMean { get; set; } |
Gets or sets the mean the log-mel targets are standardized with (LJSpeech: −5.522). |
double MelStd { get; set; } |
Gets or sets the standard deviation the log-mel targets are standardized with (LJSpeech: 2.063). |
int DecoderKernelSize { get; set; } |
Gets or sets the kernel of the decoder’s convolutions (4). |
int EncoderKernelSize { get; set; } |
Gets or sets the kernel of the encoder’s convolutions (4). |
int[] DecoderDilations { get; set; } |
Gets or sets the dilation of each decoder residual block (two convolutions each). |
int[] DurationPredictorKernelSizes { get; set; } |
Gets or sets the kernel of each duration predictor block (4, 3, 1). |
int[] EncoderDilations { get; set; } |
Gets or sets the dilation of each encoder residual block (two convolutions each). |