SpiritLmStyleTokenizer
Enums · AiDotNet.TextToSpeech.MultiModal
Where SpiritLM Expressive’s style tokens come from.
Fields
| Field | Summary |
|---|---|
SpiritLmStyleTokenizer ReleaseClassifier |
The release’s wav2vec 2.0 style classifier (style_encoder_w2v2), one class a second; the released language model’s style tokens. |
SpiritLmStyleTokenizer SpeechProp |
The paper’s (§3.2): SONAR Expressive’s SpeechProp features of the 25 Hz EnCodec units, pooled per second, fine-tuned to predict Expresso’s styles and clustered by a 100-centroid k-means. |
int value__ |