FishSpeechV15Options
Options & Configuration · AiDotNet.TextToSpeech.CodecBased
Options for FishSpeechV15.
Inheritance: ModelOptions → TtsModelOptions → FishSpeechFamilyOptions → FishSpeechV15Options
For Beginners
These settings size Fish Speech 1.5 and control how it speaks and trains; the defaults reproduce the released model.
How It Works
The defaults are Fish Speech 1.5’s (fishaudio/fish-speech-1.5): a Dual-AR transformer of 24 slow and 4 fast
layers, 1024 wide, 16 query and 2 key-value heads, a 102,048-token tiktoken vocabulary whose last 1,024 tokens are the
semantic tokens, an 8,192-token context; Firefly-GAN at 44.1 kHz with 8 codebooks of 1,024 codes at about 21 frames a
second. Training follows the paper’s recipe (Paper); the release’s fine-tuning
recipe is ApplyRecipe’s other choice.
Constructors
| Constructor | Summary |
|---|---|
FishSpeechV15Options() |
Creates the released Fish Speech 1.5 configuration. |