MegaTTSOptions
Options & Configuration · AiDotNet.TextToSpeech.Latest
Options for MegaTTS. The sizes are the paper’s (Jiang et al. 2023, Table 7); what it leaves
unstated comes from the authors’ NATSpeech and SyntaSpeech code.
Inheritance: ModelOptions → TtsModelOptions → MegaTTSOptions
For Beginners
Mega-TTS speaks in a voice from a few seconds of it, copying both how the person sounds and how they tend to speak. These options size its parts; train it on recordings with phoneme durations.
Constructors
| Constructor | Summary |
|---|---|
MegaTTSOptions() |
Creates an instance with the default settings. |
Properties
| Property | Summary |
|---|---|
MegaTtsLanguage Language { get; set; } |
The text front end (English). |
MegaTtsStage Stage { get; set; } |
The model MegaTTS trains. |
double AdversarialWeight { get; set; } |
The adversarial loss weight (SyntaSpeech’s 0.05). |
double Beta1 { get; set; } |
Adam’s β1 (0.9), β2 (0.98) and ε (1e-9), on the Transformer schedule. |
double Beta2 { get; set; } |
See Beta1. |
double DiscriminatorDecay { get; set; } |
See DiscriminatorLearningRate. |
double DiscriminatorEpsilon { get; set; } |
See DiscriminatorLearningRate. |
double DiscriminatorLearningRate { get; set; } |
The discriminator’s learning rate (1e-4), ε (1e-6), and its step decay (halved every 40000 steps), as SyntaSpeech trains it. |
double DurationWeight { get; set; } |
The phoneme-duration loss weight (NATSpeech’s 0.1); word and sentence durations weigh 1. |
double Epsilon { get; set; } |
See Beta1. |
double MaxGradientNorm { get; set; } |
The gradient norm every step clips to (NATSpeech’s clip_grad_norm of 1); 0 turns clipping off. |
double MelMaxFrequency { get; set; } |
The mel filterbank’s highest frequency (NATSpeech’s 7600 Hz). |
double MelMinFrequency { get; set; } |
The mel filterbank’s lowest frequency (NATSpeech’s 80 Hz). |
double ScheduleFactor { get; set; } |
The Transformer schedule’s scale (1: Vaswani et al.’s rate, d^−½ · min(t^−½, t · warmup^−1.5)). |
double SentenceDurationWeight { get; set; } |
See WordDurationWeight. |
double WordDurationWeight { get; set; } |
The word- and sentence-duration loss weights (NATSpeech’s 1 and 1). |
int CodeDimension { get; set; } |
See Codes. |
int Codes { get; set; } |
VQ codebook entries (2048) and their width (256). |
int ContentHeads { get; set; } |
See ContentLayers. |
int ContentLayers { get; set; } |
Content encoder layers (4) and heads (2), feed-forward kernel 5. |
int ContextSentences { get; set; } |
The sentences of one speaker the P-LLM reads as one sequence in training (Table 7’s 7 contextual sentences); see TrainProsodyModel. |
int ConvBlocks { get; set; } |
Convolution blocks of the prosody encoder’s two stacks, the timbre encoder and the mel decoder (5). |
int DiscriminatorDecaySteps { get; set; } |
See DiscriminatorLearningRate. |
int DiscriminatorHidden { get; set; } |
See DiscriminatorWindows. |
int DurationKernel { get; set; } |
See DurationLayers. |
int DurationLayers { get; set; } |
The duration predictor’s convolutions (3) and their kernel (NATSpeech’s 3). |
int LmHeads { get; set; } |
See LmLayers. |
int LmHidden { get; set; } |
See LmLayers. |
int LmLayers { get; set; } |
The prosody language model’s layers (8), width (512) and heads (8), feed-forward kernel 5. |
int NumDecoderLayers { get; set; } |
Gets or sets the convolutional depth shared by the acoustic decoder and encoders. |
int NumEncoderLayers { get; set; } |
Gets or sets the FastSpeech content encoder depth. |
int NumHeads { get; set; } |
Gets or sets the FastSpeech content encoder attention head count. |
int ProsodyBins { get; set; } |
The low mel bins the prosody encoder reads (the first 20). |
int SamplingSeed { get; set; } |
The seed of the prosody sampling. |
int TimbreKernel { get; set; } |
The timbre encoder’s kernel (31). |
int TopK { get; set; } |
Top-k sampling of the prosody codes (5). |
int TrainingSeed { get; set; } |
The seed of training’s draws (the discriminator windows). |
int VocoderInitialChannels { get; set; } |
HiFi-GAN V1’s first width (512), upsampling rates and kernels, and residual-block kernels and dilations (Kong et al. |
int WarmupSteps { get; set; } |
The Transformer schedule’s warm-up steps (4000). |
int WindowSize { get; set; } |
The analysis window, in samples (NATSpeech’s 1024). |
int[] DiscriminatorWindows { get; set; } |
The discriminator’s windows (32, 64, 128 frames) and width (192). |
int[] VocoderKernels { get; set; } |
See VocoderInitialChannels. |
int[] VocoderRates { get; set; } |
See VocoderInitialChannels. |
int[] VocoderResBlockKernels { get; set; } |
See VocoderInitialChannels. |
int[][] VocoderResBlockDilations { get; set; } |
See VocoderInitialChannels. |
string[] PhoneSet { get; set; } |
The phone set (NATSpeech’s ph_set.json): its sorted phones follow the three reserved tokens. |