BarkStageOptions
Options & Configuration · AiDotNet.TextToSpeech.CodecBased
Describes one of Bark’s transformer stages.
How It Works
Bark is not one language model. It uses a causal semantic model, a causal coarse-acoustic model, and a bidirectional fine-acoustic model. Keeping their contracts separate prevents a convenient-looking shared setting from silently producing a checkpoint-incompatible network.
Constructors
| Constructor | Summary |
|---|---|
BarkStageOptions() |
Properties
| Property | Summary |
|---|---|
bool IsCausal { get; set; } |
Whether the stage is causal. |
double Dropout { get; set; } |
Dropout probability used while training. |
int BlockSize { get; set; } |
Maximum sequence length accepted by the stage. |
int FeedForwardSize { get; set; } |
Feed-forward expansion width. |
int HiddenSize { get; set; } |
Transformer hidden width. |
int InputVocabularySize { get; set; } |
Input vocabulary width. |
int NumberOfHeads { get; set; } |
Number of self-attention heads. |
int NumberOfLayers { get; set; } |
Transformer depth. |
int OutputVocabularySize { get; set; } |
Output vocabulary width. |